By Ira Bodnar, Ryze AI · Updated · Connector tested November 13, 2026
Quick answer
The main way to connect OpenClaw to PostHog is the Ryze AI connector. Save a PostHog personal API key and its host in the PostHog card in Ryze AI, then save https://connector.get-ryze.ai/mcp in OpenClaw as a streamable-http server with auth set to oauth. Use openclaw mcp set while the Gateway is stopped, or Add server under Settings > MCP in the Control UI, and sign in with openclaw mcp login ryze. OpenClaw then lists 3 PostHog tools: one that reads and runs HogQL, one that writes and one that deletes. Changes work only if the key you created carries write scopes.
- In PostHog create a personal API key with read access to the project, including the Query and Insight scopes
- In Ryze AI open Integrations, click Connect on PostHog, paste the key, enter the host and pick the project
- Stop the Gateway and run openclaw mcp set ryze with the connector URL, or add it under Settings > MCP while the Gateway runs
- Run openclaw mcp login ryze, click Allow connection on the Ryze AI page and start the Gateway again
OpenClaw to PostHog at a glance
| Fact | Detail |
|---|---|
| Main way to connect | The Ryze AI connector, saved in OpenClaw as one remote MCP server named ryze |
| Server address | https://connector.get-ryze.ai/mcp, with transport set to streamable-http |
| Sign-in | OAuth to Ryze AI with openclaw mcp login ryze, or Sign in under Settings > MCP for an administrator. A Ryze AI API key in a bearer header is the other route |
| What PostHog needs | A phx_ personal API key with read access to the project and the Query and Insight scopes, plus your host (US Cloud, EU Cloud or self-hosted) |
| PostHog tools | 3: posthog__read, posthog__write and posthog__delete |
| Can it change PostHog | Only when the key carries write scopes. With read scopes alone, PostHog rejects every change |
| Projects | One PostHog project per Ryze AI workspace, picked after you connect |
| OpenClaw plan | None to buy. OpenClaw is MIT licensed and self-hosted, and its docs say there is no paid tier |
| Ryze AI price | Paid Ads is $89 per month with a 7-day free trial. Connecting is free; the PostHog tools need an active plan or trial |
| PostHog's own option for OpenClaw | A free hosted MCP server at https://mcp.posthog.com/mcp. PostHog documents setup for other clients and does not name OpenClaw |
| Last tested | The connector's sign-in answer was checked on November 13, 2026. No PostHog call and no step inside OpenClaw was run |
Before you start
- OpenClaw. It is an open-source project under the MIT license that you host yourself, and its docs say it has no paid tier. You need a running Gateway and one of three things: the openclaw command on the Gateway machine, administrator access to the Control UI, or write access to the Gateway's openclaw.json.
- A PostHog personal API key. It starts with phx_ and needs read access to the project, including the Query and Insight scopes. PostHog shows the key value once, when you create it, and each user can hold up to 10 keys. Add write scopes only for the objects the agent should change. You also need your host: https://us.posthog.com, https://eu.posthog.com or your own self-hosted address.
- A Ryze AI workspace on a plan or free trial. Paid Ads is $89 per month with a 7-day free trial, and SEO Starter is $129 per month with a 3-day free trial. Adding the connector is free; the PostHog tools need an active plan or trial.
- One project per workspace. A Ryze AI workspace holds one PostHog connection with one project. For a second project, use another workspace.
- A note on testing. Ryze AI has no help-center page for OpenClaw and has not tested the connector there. The OpenClaw steps follow OpenClaw's own documentation for remote servers that use OAuth, as read on November 15, 2026.
How to connect OpenClaw to PostHog
The key is made in PostHog, the PostHog connection is saved in Ryze AI, and the rest happens in OpenClaw. The connector URL is the same for every user.
Create a personal API key in PostHog
In PostHog open Settings, Personal API keys and click Create key. Grant read access to the project with the Query and Insight scopes, add write scopes only for objects the agent should change, and copy the phx_ key before you close the dialog.
Connect PostHog in Ryze AI
Create a Ryze AI account or sign in, open Integrations and click Connect on the PostHog card. Paste the key, enter the Host and click Continue, then pick the Project and click Save. You can also ask the Ryze AI agent in Claude or ChatGPT to connect PostHog: it sends a one-time link, valid once and for 24 hours, to a page where you paste the key and enter the host.
Register the connector in OpenClaw
Stop the Gateway first with openclaw gateway stop: since OpenClaw 2026.10.10 (November 3, 2026) the CLI holds back MCP changes while a Gateway owns the state directory. Then run: openclaw mcp set ryze '{"url":"https://connector.get-ryze.ai/mcp","transport":"streamable-http","auth":"oauth"}'. That stores a server named ryze. The same entry can go under mcp.servers in ~/.openclaw/openclaw.json, or be added with Add server under Settings > MCP in the Control UI while the Gateway keeps running.
openclaw gateway stop openclaw mcp set ryze '{"url":"https://connector.get-ryze.ai/mcp","transport":"streamable-http","auth":"oauth"}' openclaw mcp login ryzeSign in from the OpenClaw CLI
With the Gateway still stopped, run: openclaw mcp login ryze. A browser opens on the Ryze AI page. Sign in if asked, choose the workspace that holds PostHog and click Allow connection, then start the Gateway again. An administrator can skip the stop and click Sign in for the ryze server under Settings > MCP.
Finish the sign-in on a remote machine if needed
If the browser runs on a different machine from OpenClaw, approve the sign-in there and give the code it returns to the CLI with the --code option of openclaw mcp login. If the callback address carries an iss value, pass it with --issuer.
Ask your first question
Once credentials exist, OpenClaw adds the server to the agent runtime. If you added the connector before PostHog was connected, remove the entry and add it again so the tool list refreshes. Then ask: Run a HogQL query that counts $pageview events per day for the last 7 days, with a LIMIT of 100.
The Ryze AI help center article for PostHog covers the PostHog side of the connection. OpenClaw's own pages on the MCP registry commands and transports and OAuth cover the OpenClaw side.
Sources: OpenClaw: connect MCP servers, OpenClaw MCP registry commands, OpenClaw transports and OAuth (checked November 15, 2026).
CLI, Control UI, openclaw.json and API key
OpenClaw CLI
openclaw gateway stop
openclaw mcp set ryze '{"url":"https://connector.get-ryze.ai/mcp","transport":"streamable-http","auth":"oauth"}'
openclaw mcp login ryzeset writes the server definition and login runs the OAuth flow, with the credentials kept on the Gateway host. Start the Gateway again afterwards. The flag form is openclaw mcp add ryze --url https://connector.get-ryze.ai/mcp --transport streamable-http --auth oauth. The commands openclaw mcp show ryze, list and status print what is saved.
Control UI: Settings > MCP
Open the MCP page of the Control UI (/settings/mcp, with /mcp as an alias), choose Add server under Configured servers, name it ryze, pick Streamable HTTP and paste the URL. OpenClaw's docs put OAuth settings in the scoped config editor lower on that page, so set auth: "oauth" there and click Sign in. The page needs administrator access, and Sign in works only when the Control UI is opened on the Gateway's loopback address, its own HTTPS address or its published Tailscale address. In a chat, + > Plugins > Add MCP server saves a server for This session or Everywhere.
openclaw.json
{
mcp: {
servers: {
ryze: {
url: "https://connector.get-ryze.ai/mcp",
transport: "streamable-http",
auth: "oauth",
},
},
},
}Edit ~/.openclaw/openclaw.json, which is JSON5, so comments and trailing commas are allowed. The Gateway watches the file and applies edits. A key it does not recognize stops it from starting, so keep the entry under mcp.servers. Sign in afterwards with openclaw mcp login ryze on a stopped Gateway, or with Sign in under Settings > MCP.
A Ryze AI API key instead of OAuth
ryze: {
url: "https://connector.get-ryze.ai/mcp",
transport: "streamable-http",
headers: { Authorization: "Bearer ${RYZE_API_KEY}" },
}Drop auth: "oauth" and send the key as a bearer header, because OpenClaw ignores static Authorization headers while auth: "oauth" is on. This key belongs to Ryze AI and is separate from the PostHog phx_ key. Keep it in an environment variable: a ${RYZE_API_KEY} reference does not trigger the literal-value warning of openclaw doctor when the variable is set.
Transport and timeouts
A remote entry takes the transport streamable-http or sse, and OpenClaw reads a URL entry with no transport as sse, so set streamable-http for this address. A request times out after 60 seconds by default, and the tool list at session start gets 10 seconds when requestTimeoutMs is not set.
OpenClaw and MCP: the facts in one table
| Question | Answer from OpenClaw's docs |
|---|---|
| Where a server is added | Settings > MCP > Add server in the Control UI, + > Plugins > Add MCP server in a chat, openclaw mcp add or set in a terminal, or mcp.servers in ~/.openclaw/openclaw.json |
| Config file | ~/.openclaw/openclaw.json in JSON5. The Gateway applies edits on its own; an unknown key blocks startup |
| Transports | stdio for local commands, plus SSE and Streamable HTTP for addresses. type: "http" is rewritten to transport: "streamable-http" by openclaw mcp set and openclaw doctor --fix |
| Sign-in | auth: "oauth" on an HTTP entry, then openclaw mcp login with the Gateway stopped, or Sign in under Settings > MCP for an administrator |
| Whose account | By default one shared credential that the operator manages. oauth.identity: "per-requester" gives each sender a sign-in of their own and needs gateway.publicOrigin set to the Gateway's HTTPS origin |
| Approval | MCP tools follow the same tool profiles and tool policy as other tools. The coding and messaging profiles include them, minimal hides them, and tools.deny: ["bundle-mcp"] turns them off. Per-call prompts are documented for Codex runs only |
| Tool names | The agent sees <server>__<tool>. openclaw mcp doctor ryze --probe opens a live connection and lists the tools |
| Plan | No paid tier; MIT licensed and self-hosted |
| Admin controls | Adding a server from a chat and signing in from Settings need administrator access. enabled: false keeps an entry and hides its tools |
Sources: OpenClaw: connect MCP servers, MCP registry commands, transports and OAuth, MCP config, Gateway configuration, MCP in the Control UI, OpenClaw docs home (checked November 15, 2026).
Other clients
The same URL works in other MCP clients. PostHog is connected once in Ryze AI, so the key and project stay as they are.
Claude Code
Claude Code registers the same URL with one claude mcp add command.
ChatGPT and Claude
Both add the URL as a custom connector, with plan limits that differ from OpenClaw's.
Permissions and access
Four things decide what OpenClaw can reach in PostHog.
- Which key and host
- Ryze AI stores the phx_ key and the host you enter. A host that does not match where the account lives, or a key without the Query and Insight scopes, is rejected.
- Which project
- One connection holds one PostHog project. PostHog is not on the Ryze AI list of platforms that support several accounts. To switch, click View on the card, disconnect with the trash icon and connect again, or use another workspace for the other project.
- Read or write
- What the agent can change is limited by the scopes on the key. A key with read scopes only keeps OpenClaw read-only, because PostHog rejects any change.
- One shared sign-in
- By default OpenClaw's OAuth credential is shared and operator-managed, so everyone who talks to that agent works through the one Ryze AI sign-in and the one PostHog key behind it. Per-requester sign-in needs oauth.identity set to per-requester and gateway.publicOrigin set to an HTTPS origin.
What the connection is allowed to do
Two sign-ins stand between OpenClaw and your analytics. OpenClaw signs in to the Ryze AI connector, and Ryze AI holds your PostHog personal API key. The key acts as the person who made it, within the scopes and projects set on it. Each PostHog scope has a read level and a write level, and write includes read.
| PostHog scope | What PostHog says it covers | On the key Ryze AI asks for |
|---|---|---|
| query:read | SQL and insight queries on the project's data, plus event, session and web vitals lookups. There is no write level | Yes, as the Query scope |
| insight:read, insight:write | Insights with their queries, results and explanations | Read, as the Insight scope. Write only to create or edit insights |
| dashboard:read, dashboard:write | Dashboards with their tiles, text cards, widgets and layout | Not named. Add read if a dashboard request fails with a 403 that names it |
| feature_flag:read, feature_flag:write | Flags with their release conditions, variants, payloads and rollout history | Not named. Same rule as dashboards |
| cohort:read, cohort:write | Cohorts with their filters, static member lists and calculation history | Not named. Add read if a cohort list fails with a 403; write is needed to create or delete a cohort |
| person:read, person:write | Persons and their properties | Not named. Add read if a request for persons fails with a 403 |
| event_definition:read, event_definition:write | Event definitions, schemas and property groups | Not named. Add read if listing event definitions fails with a 403 |
Not included by default: any write scope, a second project in the same workspace, and the MCP Server key preset, which PostHog made for its own MCP server and CLI. PostHog's scopes page says a request that needs a scope the key lacks fails with a 403 that names the scope. Ryze AI does not list session recordings, experiments, surveys or error tracking among what its tools cover, so this guide makes no claim about them.
A key can cover the whole organization, chosen projects or every project, and its scopes can be changed later. Organization admins and owners can see each member's key with its scopes and last use, and GitHub secret scanning rolls an exposed phx_ key and emails its owner.
Sources: PostHog API scopes, Personal API keys, PostHog API overview (checked November 14, 2026).
Removing access
To cut OpenClaw off, stop the Gateway and run openclaw mcp logout ryze, which deletes the saved credentials but keeps the entry, or openclaw mcp unset ryze, which removes the entry. enabled: false on the entry hides its tools without deleting anything. To end the PostHog access itself, click View on the PostHog card in Ryze AI and disconnect, or delete the personal API key in PostHog.
What OpenClaw can do in PostHog
The Ryze AI connector exposes 3 PostHog tools: 1 read, 1 write and 1 delete. They run against the one project you picked. The names come from Ryze AI's facts file; they were not listed from a live connection for this guide.
| Tool | Type | What it does |
|---|---|---|
| posthog__read | Read | Reads projects, insights, dashboards, feature flags, cohorts, persons and event definitions, and runs HogQL analytics queries. |
| posthog__write | Write | Creates or updates PostHog objects such as insights, dashboards, feature flags and cohorts. |
| posthog__delete | Delete | Deletes a PostHog object by path, for example a cohort. The deletion is irreversible. |
Keeping changes in check
In its built-in runtime OpenClaw does not document a prompt before each MCP call; those are documented for Codex runs only. MCP tools follow the usual tool profiles and policy, and inside a chat, + > Plugins > Tool access lets you deny single tools for that session. The standing way to keep the agent from changing PostHog is a read-only key, because PostHog rejects any change without write scopes. As a second layer, a toolFilter with an exclude list on the entry hides the write and delete tools. Entries take exact names or simple * globs; run openclaw mcp doctor ryze --probe first to see the names OpenClaw reports. In the Ryze AI chat the Tool approvals picker starts on Skip all approvals, so the agent runs PostHog tools on its own; switch to Approve write & delete or Manual approve. Scheduled tasks in Ryze AI run without an approval step.
ryze: {
url: "https://connector.get-ryze.ai/mcp",
transport: "streamable-http",
auth: "oauth",
toolFilter: {
exclude: ["*posthog__write", "*posthog__delete"],
},
}Limits in numbers
Ryze AI adds one rule of its own: a HogQL query needs a time window and a row limit of 1,000 or fewer. Every other number below belongs to PostHog or OpenClaw, counted the way that vendor counts it. Which PostHog limit a given Ryze AI call draws on was not tested.
| Limit | Figure | Set by |
|---|---|---|
| HogQL through Ryze AI | A time window and a row limit of 1,000 at most | Ryze AI facts |
| Analytics endpoints | 240 requests per minute and 1,200 per hour for each personal API key, for example to calculate insights or retrieve persons | PostHog |
| Query endpoint | 2,400 per hour on the API overview. The queries page counts it per project: 240 per minute, 3 queries running at once, 60 threads per query and 10 seconds of execution time. A queued query may wait up to 30 seconds | PostHog |
| Hourly read budget | Queries made with a personal API key draw from a budget of bytes read per project. When it is spent the answer is 429 with api_queries_budget_exceeded and a Retry-After header, and cached results still work. PostHog gives no byte figure | PostHog |
| Other read, create, update and delete calls | 480 per minute and 4,800 per hour for each personal API key | PostHog |
| Who the limits count against | Most count per personal API key, not per team. Requests made with OAuth or a logged-in session do not count against them, and PostHog says it is not offering higher limits | PostHog |
| Rows in one query | 100 by default, up to 50,000 with your own LIMIT. OFFSET answers HTTP 400 for personal API keys, so page with keyset pagination on timestamp | PostHog |
| Query cache | API queries are cached by default, and a response says so with is_cached and last_refresh | PostHog |
| Personal API keys per user | Up to 10 | PostHog |
| Dynamic cohorts | Updated once every 24 hours | PostHog |
| OpenClaw request time | 60 seconds per request, and 10 seconds for the tool list at session start unless requestTimeoutMs is set | OpenClaw |
PostHog's own MCP server has a separate cap of 480 requests per minute and 4,800 per hour for each credential, with a 429 and a Retry-After header past it. That cap applies to mcp.posthog.com and not to the Ryze AI connector. The PostHog pages read gave no figure for the delay between an event being sent and being queryable, so this guide gives none.
Sources: PostHog API overview, PostHog queries API, Personal API keys, PostHog cohorts, PostHog MCP FAQ (checked November 14, 2026); OpenClaw MCP config, OpenClaw: connect MCP servers (checked November 15, 2026).
First 3 prompts to run
Each answer comes from your own project, so this page describes what comes back instead of showing sample numbers.
1. A bounded HogQL query
Run a HogQL query on my PostHog project that counts $pageview events per day for the last 7 days, with a LIMIT of 100.
Tools used: posthog__read
What comes back: A table with one row per day and its pageview count. Keep the time window and the row limit in the prompt, because Ryze AI expects both.
2. Traffic by campaign
Break last month's sessions down by UTM source with a HogQL query over sessions, limited to 50 rows.
Tools used: posthog__read
What comes back: One row per UTM source with a session count, for the sources PostHog has recorded. Spell out the field you want grouped in the prompt.
3. A saved funnel
Open my saved signup funnel insight and tell me how many people reach each step.
Tools used: posthog__read
What comes back: The step counts of that insight. This needs the Insight and Query scopes on the key.
To create something, say what and where in plain words, for example a trends insight for landing page signups on a new campaign dashboard. That needs insight:write and dashboard:write on the key. Deleting is a separate tool, so name the object you want removed. The Ryze AI help center describes deleting insights, dashboards, feature flags and cohorts when explicitly asked.
10 jobs to hand to OpenClaw
Ten jobs to give an OpenClaw agent, each with the Ryze AI tool and the PostHog scope behind it. None was run for this guide. The last column assumes the key Ryze AI asks for: Query and Insight scopes and no write scopes.
| Job | What to ask OpenClaw | Tool | PostHog scope | Works with that key |
|---|---|---|---|---|
| Daily pageviews | Count pageviews per day for the last 7 days with HogQL, capped at 1,000 rows. | posthog__read | query:read | Yes |
| Traffic by campaign | Split last month's sessions by UTM source to see which campaigns sent visitors. | posthog__read | query:read | Yes |
| Funnel readout | Read my saved signup funnel and report how many people reach each step. | posthog__read | insight:read, query:read | Yes |
| Dashboard names | List the dashboards in the project by name. | posthog__read | dashboard:read | Not confirmed. It works when the key also has dashboard:read |
| Flag check before a launch | List the feature flags and their current rollout settings. | posthog__read | feature_flag:read | Not confirmed. It works when the key also has feature_flag:read |
| Cohort review | List the cohorts and what defines each one. | posthog__read | cohort:read | Not confirmed. It works when the key also has cohort:read |
| Event inventory | List the event definitions so I can see which events exist before building a report. | posthog__read | event_definition:read | Not confirmed. It works when the key also has event_definition:read |
| New campaign dashboard | Create a trends insight for landing page signups and add it to a new campaign dashboard. | posthog__write | insight:write, dashboard:write | No. Add both write scopes first |
| Campaign cohort | Create a cohort of people who arrived from one campaign. | posthog__write | cohort:write | No. It needs cohort:write |
| Remove a test cohort | Delete the old test cohort I name here. This cannot be undone. | posthog__delete | cohort:write | No. It needs cohort:write and a request that names the cohort |
Some jobs have no Ryze AI tool: session recordings, experiments, surveys and error tracking are not listed among what the PostHog tools cover, and a second project needs another Ryze AI workspace.
5 ways to connect OpenClaw to PostHog
Five routes to PostHog are described by their makers in terms that fit OpenClaw's remote-server support, and three of them name OpenClaw. This guide treats the first, the Ryze AI connector, as the main way: one sign-in on the PostHog side and one remote entry in OpenClaw. The pages below were read on November 14, 2026, and nothing from them was installed inside OpenClaw.
| Way | Runs where | Setup for OpenClaw | What it reaches in PostHog | Price |
|---|---|---|---|---|
| Ryze AI connector (the main way) | Hosted by Ryze AI, one address for every user | A streamable-http entry named ryze with auth oauth, from Settings > MCP, openclaw mcp set or openclaw.json, then one sign-in. A Ryze AI API key in a bearer header can stand in for the sign-in | 3 tools on one project: insights, dashboards, flags, cohorts, persons, event definitions and HogQL. Changes need write scopes on the key | Paid Ads $89 per month with a 7-day free trial, or SEO Starter $129 per month with a 3-day free trial |
| PostHog's hosted MCP server | Hosted by PostHog at https://mcp.posthog.com/mcp, routed to your US or EU region at sign-in | PostHog documents setup for Claude Code, Claude Desktop, Cursor, Codex, VS Code, Windsurf and Zed, and does not name OpenClaw. OpenClaw can save a remote address with auth oauth, and PostHog says a client without OAuth can send a personal API key made with the MCP Server preset as a bearer header. Neither was tried in OpenClaw | 1,284 tools in 78 categories, read and write, with a read-only switch (?readonly=true) and tool filters. One connection has one active organization and project | Free to connect and call. PostHog says some tools use LLMs internally and may count as PostHog AI spend |
| Composio | Hosted by Composio at https://connect.composio.dev/mcp | Composio's PostHog page lists OpenClaw among the assistants it works with. Setup is a Streamable HTTP address and a login to Composio, and PostHog is linked with your own API key. No OpenClaw-specific steps were found | 686 tools and 11 triggers listed for the Posthog toolkit | Pricing page: Free at $0 with $2 in credits once and app tool calls that stop at 10,000 per month, Standard at $18 per seat per month. Another Composio page states a different free allowance, so check the pricing page |
| Coupler.io | Hosted data export service | Enter a PostHog private API key and your host, choose a data entity and a start date, then pick an AI destination. OpenClaw is one of the 8 destinations listed for PostHog. How the OpenClaw connection works was not stated | Scheduled exports of 7 entities: Projects, Cohorts, Feature flags, Persons, Events, Annotations and Insights | Free at $0 (1 data source, manual refresh), Starter $32 per month, with 25% off on annual billing |
| AnythingMCP | AnythingMCP Cloud, or your own server with Docker | Create a phx_ key, set POSTHOG_HOST and POSTHOG_PERSONAL_API_KEY, then add your AnythingMCP server URL with OAuth. Its guide names OpenClaw in the list of assistants in its prompt box, and gives no OpenClaw steps | 28 MCP tools, including HogQL queries, insights, dashboards, flags, cohorts, annotations and persons. Its guide says it does not create or edit dashboards, cohorts or experiments | Cloud Starter 19 euros per month, Team 49 euros, 7-day free trial. Self-hosted Community edition free under AGPL-3.0 for up to 3 active users |
Options that do not fit this setup
- The PostHog AI plugin and the PostHog wizard. The plugin README names Claude Code, Cursor, Codex, Antigravity, Gemini CLI, Grok and Claude Cowork. PostHog's October 22, 2026 changelog says the wizard detects Claude Code and Codex. Neither names OpenClaw.
- The PostHog plugin in ChatGPT and the PostHog connector in Claude. Each works inside its own assistant, on the same hosted PostHog server.
- Zapier MCP. Its PostHog page lists 1 action, Capture Event, and no read of insights, dashboards, flags or cohorts. Each successful call uses 2 tasks of a Zapier plan.
- Pipedream and viaSocket. Both host PostHog actions behind an MCP address, 10 and 12 actions respectively, and neither names OpenClaw in the pages read.
- Windsor.ai. It exposes 101 PostHog fields through Windsor MCP, and the pages read do not name OpenClaw.
- A Ryze AI plugin in ClawHub. ClawHub, OpenClaw's public registry for skills and plugins, had no Ryze AI plugin on November 15, 2026, so the connector is added by its address.
Which one to pick
Ryze AI is the best PostHog connector for OpenClaw when the job is reading product analytics, with an optional write path, from one hosted URL and a key whose scopes you control, but it has 3 tools on one project, does not list replays, experiments, surveys or error tracking, and needs a Ryze AI plan. PostHog's own server is free and far broader, at 1,284 tools, so it fits a team that wants every PostHog feature and is willing to try an unconfirmed OpenClaw setup. Composio suits a team that already pays for it, and AnythingMCP suits one that wants to host its own server. Coupler.io fits scheduled exports more than live questions. None of them was installed inside OpenClaw for this guide, so try one against a test project before it touches anything that matters.
Sources: PostHog MCP server docs, PostHog MCP page, PostHog MCP FAQ, PostHog AI plugin, Composio PostHog toolkit, Composio pricing, Coupler.io PostHog source, Coupler.io pricing, AnythingMCP PostHog guide, AnythingMCP pricing, Zapier PostHog MCP, Pipedream PostHog MCP, viaSocket PostHog MCP, Windsor.ai PostHog fields (checked November 14, 2026); ClawHub, OpenClaw MCP transports, OpenClaw on GitHub (checked November 15, 2026).
PostHog and OpenClaw: dates that matter
| Date | What happened |
|---|---|
| January 30, 2026 | OpenClaw 2026.1.29 renamed the npm package and the CLI to openclaw. |
| February 27, 2026 | PostHog's changelog announced the PostHog connector for Claude, and added cohort management through its MCP server the same day. |
| March 31, 2026 | OpenClaw 2026.3.31 added remote HTTP and SSE servers by URL under mcp.servers, with auth headers and an optional streamable-http transport. |
| May 4, 2026 | PostHog added explicit MCP mode selection with the x-posthog-mcp-mode header or the mode query parameter. |
| August 3 to October 15, 2026 | PostHog's changelog lists new MCP tools in nine batches, from 5 to 44 tools each. |
| August 31, 2026 | OpenClaw 2026.8.1, also called OpenClaw 2.0, added servers from the chat composer and per-requester OAuth for shared servers. |
| September 23, 2026 | OpenClaw 2026.9.6 let administrators sign in to MCP connectors from Settings. |
| October 1, 2026 | PostHog made MCP connections longer lived for clients with non-rotating refresh tokens, which previously asked for a new authorization after roughly 67 days. |
| October 22, 2026 | PostHog's wizard began detecting Claude Code and Codex and installing the MCP connection. A separate flow handles browser and account-based clients such as ChatGPT and Claude. |
| November 3, 2026 | OpenClaw 2026.10.10 began protecting MCP configuration while the Gateway runs. The commands openclaw mcp set and login now need the Gateway stopped. |
| November 5, 2026 | New PostHog event retention periods apply to organizations that start a paid plan on or after this date. |
| November 10, 2026 | OpenClaw 2026.11.3 came out. On November 15, 2026 it was the newest stable release, and its notes list no MCP change. |
Sources: PostHog changelog 2026, PostHog events retention (checked November 14, 2026); OpenClaw releases, changelog 2026.3.31, changelog 2026.8.1, release 2026.9.6, release 2026.10.10 (checked November 15, 2026).
How this guide was tested
Three dates stand behind this guide. The connector's sign-in answer was checked on November 13, 2026, the PostHog pages were read on November 14, 2026, and the OpenClaw steps were compared with OpenClaw's documentation on November 15, 2026. No call to PostHog was run and nothing was run inside OpenClaw.
| What was run | What came back |
|---|---|
| The connector address without a sign-in, November 13, 2026 | HTTP 401. OpenClaw's docs say a 401 at startup is logged as authentication refused (401); that line was not seen, because OpenClaw was not run. |
| PostHog docs, API reference, scopes and pricing pages, November 14, 2026 | Read, and every figure on this page copied from them with the source linked. |
| PostHog's MCP page, changelog and the listings of Composio, Coupler.io and AnythingMCP, November 14, 2026 | Read, to describe the other ways to connect fairly. Nothing was installed. |
| OpenClaw docs, changelog files and GitHub releases, November 15, 2026 | Read, and each release date matched against the published date of the GitHub release. |
Not run for this page: every step inside OpenClaw and every call to PostHog. No Gateway was installed or started, no server was saved and no sign-in was tried, so openclaw mcp set, openclaw mcp login, Settings > MCP and the openclaw.json entry follow OpenClaw's documentation as read on November 15, 2026. No PostHog test project was connected, so posthog__read, posthog__write and posthog__delete were not called, and this guide quotes no result and no error message from a run.
Errors and fixes
Messages from the vendor docs
These messages come from PostHog's and OpenClaw's documentation, not from a test run for this guide. If one turns up in an OpenClaw chat or log, this is what the vendor says it means.
| Message | Meaning | What to do |
|---|---|---|
| Invalid Personal API key. (401, invalid_personal_api_key) | PostHog does not accept the personal API key. | Create a new key in PostHog, then click View on the PostHog card in Ryze AI, disconnect and connect again with the new key. |
| A 403 that names a missing scope | The request needs a scope the key does not have, for example dashboard:read or cohort:write. | Add the named scope to the key in PostHog. Scopes can be changed after the key is made. |
| 429 with api_queries_budget_exceeded and a Retry-After header | The project's hourly read budget for personal API key queries is used up. Cached results still work. | Wait for the Retry-After time, or ask for smaller time windows. |
| HTTP 400 on OFFSET | OFFSET paging on the query endpoint is rejected for personal API keys. | Ask for keyset paging on timestamp, or a narrower time window. |
| authentication refused (401) | OpenClaw's docs use this line when a Streamable HTTP server rejects startup with a 401, which is how the Ryze AI connector answers before you sign in. The pairing comes from the docs and was not seen in a run. | Run openclaw mcp login ryze with the Gateway stopped. |
| authorization-required, or insufficient_scope | OpenClaw's names for a server that still needs sign-in, or a token that lacks a scope. | Run openclaw mcp login ryze again. |
Other problems
- The PostHog tools do not appear in OpenClaw
- Until shared credentials exist, OpenClaw leaves the server out of the agent runtime instead of failing the turn. Run openclaw mcp login ryze with the Gateway stopped, then check with openclaw mcp status or openclaw mcp doctor ryze --probe. If you added the connector before PostHog was connected in Ryze AI, remove the entry and add it again.
- The tool list is empty or times out
- The tool list at session start gets 10 seconds when requestTimeoutMs is not set. Raise requestTimeoutMs on the entry, and confirm transport is streamable-http, because OpenClaw reads a missing transport as sse.
- The sign-in page opens on the wrong machine
- Approve it in the browser, then give the code it returns to openclaw mcp login ryze with --code, and the iss value with --issuer if the callback carries one.
- The API key is ignored
- Static Authorization headers are ignored while auth: "oauth" is on. Remove auth for the key route, or remove the header and use OAuth.
- An openclaw mcp command holds back
- Since OpenClaw 2026.10.10 (November 3, 2026), commands that change saved servers, and login, need the Gateway stopped. Stop it, run the command and start it again. To skip the stop, edit openclaw.json, which the Gateway watches, or use Settings > MCP.
- The Gateway will not start after editing openclaw.json
- A key the Gateway does not know stops it. Servers belong under mcp.servers in ~/.openclaw/openclaw.json, so move any block copied from another format, such as an mcpServers block, into that shape.
- PostHog says the key or host is wrong
- Check that the host matches where the account lives: https://us.posthog.com for US Cloud, https://eu.posthog.com for EU Cloud, or your own address.
- The agent will not change anything
- A key with read scopes only keeps the agent read-only. Add the write scope for the object, such as insight:write or cohort:write, to the key in PostHog.
- A HogQL question fails or comes back cut short
- Ask for a time window and a LIMIT of 1,000 or fewer. PostHog caps a query at 10 seconds of execution time and returns 100 rows unless you set a LIMIT.
- You need a second project
- A workspace holds one PostHog project. Use another Ryze AI workspace, or disconnect and connect again with the other key.
- The one-time connection link is refused
- The link is refused while a single-account platform like PostHog is already connected in the workspace. Switch projects or keys in the Ryze AI app.
- The agent says a subscription is required
- The tools need an active Ryze AI plan or free trial. Without one the connector can only list workspaces and connect platforms.
Terms on this page
- Gateway
- OpenClaw's single process on your machine or server. It reads its server list from openclaw.json and keeps MCP sign-ins in its state directory.
- toolFilter
- A setting on a server entry that keeps (include) or hides (exclude) tools by exact name or * pattern.
- Personal API key
- A PostHog key that starts with phx_ and authenticates the private API as your own user. You choose its scopes, and you can create up to 10.
- Scope
- PostHog's unit of access. In its words, a scope grants access to one part of PostHog, such as feature flags or insights, with a read level and a write level.
- HogQL
- PostHog's SQL, a wrapper around ClickHouse SQL, used to query events, persons and sessions.
- Host
- The PostHog address your account lives on: https://us.posthog.com, https://eu.posthog.com or your own domain when self-hosted.
- Cohort
- A list of users who have something in common. Static cohorts do not change; dynamic cohorts update once every 24 hours.
Sources: OpenClaw Gateway configuration, OpenClaw MCP config (checked November 15, 2026); PostHog API scopes, PostHog SQL, PostHog cohorts (checked November 14, 2026).
Frequently asked questions
How do I connect OpenClaw to PostHog?
Create a PostHog personal API key and connect it in the PostHog card in Ryze AI with your host and project. Then stop the Gateway, run openclaw mcp set ryze with the URL https://connector.get-ryze.ai/mcp, transport streamable-http and auth oauth, and run openclaw mcp login ryze. Click Allow connection and start the Gateway again. An administrator can do the same from Settings > MCP in the Control UI without stopping the Gateway.
Is there a PostHog MCP server for OpenClaw?
Yes, the Ryze AI connector at https://connector.get-ryze.ai/mcp, which OpenClaw saves as a remote server and which brings 3 PostHog tools. PostHog also hosts a free MCP server at https://mcp.posthog.com/mcp, but its docs set up other clients and do not name OpenClaw. Composio, Coupler.io and AnythingMCP name OpenClaw in their PostHog pages.
Is there a PostHog connector for OpenClaw?
OpenClaw's MCP pages describe how to add servers and name no built-in PostHog connector. ClawHub had no Ryze AI plugin on November 15, 2026, so the Ryze AI connector is added by its URL as a remote MCP server.
Can OpenClaw change things in PostHog?
Yes, when the personal API key has write scopes. posthog__write creates or updates insights, dashboards, feature flags and cohorts, and posthog__delete removes an object by path. With read scopes only, PostHog rejects every change.
Can OpenClaw delete things in PostHog?
Yes, if the key can write them and you ask by name. posthog__delete deletes a PostHog object by path, such as a cohort, and the deletion is irreversible. A toolFilter that excludes the delete tool keeps the agent from using it.
Which PostHog scopes does the key need?
Read access to the project, including the Query and Insight scopes. Reading dashboards, flags, cohorts, persons or event definitions may also need their own read scopes, and PostHog answers a missing scope with a 403 that names it. Add write scopes only for what the agent should change.
Which PostHog route should I pick for OpenClaw?
Pick the Ryze AI connector for one hosted URL, three tools and a key you scope yourself. Pick PostHog's own server for its full tool catalog, accepting that its OpenClaw setup is not documented by PostHog. Each of the other routes asks for more setup than one remote entry and one sign-in.
Do I need a paid OpenClaw plan?
No. OpenClaw is open source (MIT license per its GitHub repository) and self-hosted, and its docs say it has no paid tier. The cost is the Ryze AI plan or trial that the PostHog tools need.
Should I use OAuth or an API key for the Ryze AI sign-in?
OAuth through openclaw mcp login is the default path here. A Ryze AI API key sent as a bearer header suits setups where no browser sign-in is practical. OpenClaw ignores static Authorization headers while auth is set to oauth, so use one method per entry. Neither replaces the PostHog personal API key, which you save in Ryze AI.
Does it work with PostHog EU Cloud or a self-hosted PostHog?
Yes. The host field takes https://us.posthog.com for US Cloud, https://eu.posthog.com for EU Cloud or a self-hosted URL.
Can I connect more than one PostHog project?
One workspace holds one PostHog connection with one project. Use another Ryze AI workspace for another project, or disconnect and connect again with a different key.
Why does the server not appear in my OpenClaw agent?
Until shared credentials exist, OpenClaw leaves the server out of the agent runtime. Stop the Gateway, run openclaw mcp login ryze, start the Gateway again and check with openclaw mcp status.
What if OpenClaw runs on a server without a browser?
Open the sign-in link in a browser on another machine, approve it there and give the returned code to the --code option of openclaw mcp login. If the callback also carries an iss value, pass it with --issuer.
How much does it cost to connect OpenClaw to PostHog?
Connecting costs nothing, but the 3 PostHog tools need an active Ryze AI plan or free trial. Paid Ads is $89 per month with a 7-day free trial, and SEO Starter is $129 per month with a 3-day free trial. Annual billing takes 30% off the first year. PostHog itself bills by its own pricing.
How do I disconnect PostHog from OpenClaw?
Stop the Gateway and run openclaw mcp logout ryze to drop the credentials, or openclaw mcp unset ryze to remove the entry. To end the PostHog access, disconnect PostHog under Integrations in Ryze AI or delete the personal API key in PostHog.