How to connect Hermes Agent to PostHog

Save one PostHog project in Ryze AI, add one ryze entry to ~/.hermes/config.yaml, run hermes mcp login ryze, then ask Hermes Agent about insights, flags, cohorts and HogQL.

Your MCP URL
https://connector.get-ryze.ai/mcp

Copies the URL for the url field of a ryze entry under mcp_servers in config.yaml. See the steps

See pricing

By Ira Bodnar, Ryze AI · Updated · Sources checked November 14, 2026

Quick answer

The main way to connect Hermes Agent to PostHog is the Ryze AI connector. Save a PostHog personal API key and your host in Ryze AI and pick the project, add a ryze entry with the URL https://connector.get-ryze.ai/mcp and auth: oauth under mcp_servers in ~/.hermes/config.yaml, then run hermes mcp login ryze and click Allow connection. Start a new session and ask about your PostHog data.

  1. In PostHog create a personal API key (phx_) with the Query and Insight scopes
  2. In Ryze AI open Integrations, click Connect on PostHog, paste the key, enter the host and pick the Project
  3. Add the ryze entry to mcp_servers in ~/.hermes/config.yaml and run hermes mcp login ryze
  4. Start a new session or run /reload-mcp, then ask your first PostHog question

Hermes Agent to PostHog at a glance

FactDetail
Main way to connectThe Ryze AI connector, a remote HTTP MCP server that Hermes Agent reads from one ryze entry under mcp_servers in ~/.hermes/config.yaml
Connector addresshttps://connector.get-ryze.ai/mcp, identical for every user
How Hermes Agent signs inOAuth with auth: oauth, completed once with hermes mcp login ryze. A Ryze AI API key sent as a bearer header from ~/.hermes/.env is the alternative
What PostHog needsA phx_ personal API key with read access to the project, including the Query and Insight scopes, plus the host: US Cloud, EU Cloud or a self-hosted URL
PostHog tools3 in total: 1 that reads, 1 that writes and 1 that deletes
Changes to the projectOnly when the key holds write scopes. With read scopes alone PostHog rejects every change. Setting trust: untrusted on the ryze entry makes Hermes Agent ask before write-capable calls
Projects per connectionOne. A second project goes in a second Ryze AI workspace
Hermes Agent planNone. Hermes Agent is open source under the MIT license per its GitHub repository and runs on your own machine or server
Ryze AI pricePaid Ads costs $89 per month after a 7-day free trial. Adding the connector costs nothing
PostHog's own optionA free hosted MCP server at https://mcp.posthog.com/mcp. PostHog's pages name Claude Code, Claude Desktop, Cursor, Codex, VS Code, Windsurf and Zed, and do not name Hermes Agent
Last testedConnector facts recorded November 8, 2026. No PostHog call was run for this guide, and the steps inside Hermes Agent were not run

Before you start

  • Hermes Agent installed. It is open source from Nous Research, licensed MIT per its GitHub repository, and self-hosted, so there is no plan tier to buy. The configuration lives in ~/.hermes/config.yaml.
  • A PostHog personal API key. It starts with phx_ and needs read access to the projects you want, including the Query and Insight scopes. Add write scopes only if Hermes Agent should change insights, dashboards, feature flags or cohorts. You also need the host: https://us.posthog.com for US Cloud, https://eu.posthog.com for EU Cloud, or your self-hosted URL.
  • A Ryze AI workspace on a plan or free trial. Paid Ads is $89 per month with a 7-day free trial; the PostHog tools also work on SEO Starter ($129 per month, 3-day free trial) and Agency ($279 per month). Connecting the connector is free, but the PostHog tools need an active plan or trial.

How to connect Hermes Agent to PostHog with OAuth

Two steps happen in PostHog and Ryze AI, the rest in Hermes Agent. The connector URL is the same for every user.

  1. Create a PostHog personal API key

    In PostHog go to Settings, Personal API keys and click Create key. Grant read access to the projects you want, with the Query and Insight scopes. Add write scopes only for objects Hermes Agent should change, then copy the phx_ key.

  2. Connect PostHog in Ryze AI

    Create a Ryze AI account or sign in, open Integrations, click Connect on the PostHog card, paste the key, enter the Host (US Cloud, EU Cloud or your self-hosted URL) and click Continue. Then pick the Project and click Save.

  3. Add the ryze entry to config.yaml

    Open ~/.hermes/config.yaml and add a ryze server under mcp_servers. Editing the file inside a running session starts an automatic MCP reload with a 30-second timeout, which is too short for a browser sign-in, so add the entry and then run the login in the next step.

    mcp_servers: ryze: url: "https://connector.get-ryze.ai/mcp" auth: oauth
  4. Sign in to the connector

    Run hermes mcp login ryze. Hermes Agent prints an authorize URL and opens your browser when it can. Sign in on the Ryze AI page, choose the workspace and click Allow connection. Hermes Agent caches the tokens at ~/.hermes/mcp-tokens/ryze.json and reuses them until a refresh fails.

  5. Load the tools

    Start a new session, or run /reload-mcp inside a running one. Hermes Agent registers each tool under the name mcp_<server>_<tool>, so the PostHog tools start with mcp_ryze_. To see what the server answers, run hermes mcp test ryze.

  6. Ask your first question

    Ask: List the saved insights in my PostHog project.

The Ryze AI help center page for PostHog covers the PostHog side of the connection, and the Ryze AI help article on API keys and the MCP endpoint covers the connector URL. The Ryze AI developers page lists the same URL.

Source for the Hermes Agent steps: Hermes Agent MCP documentation, opened November 19, 2026.

API key, tool filters and headless hosts

Hermes Agent keeps every MCP server in one file, ~/.hermes/config.yaml. These are the variations that matter for this connector.

A Ryze AI API key instead of OAuth

Create a key in Ryze AI under Workspace Settings > Profile > API keys > Create key and copy it, because it is shown once. Put RYZE_API_KEY with the rk_live_ key in ~/.hermes/.env and reference it in the header, since Hermes Agent resolves ${VAR} placeholders in headers from environment variables, which include everything in that file. Run /reload-mcp afterwards. This is a Ryze AI key and is separate from the phx_ PostHog key, which stays in Ryze AI.

mcp_servers: ryze: url: "https://connector.get-ryze.ai/mcp" headers: Authorization: "Bearer ${RYZE_API_KEY}"

Show Hermes Agent only the read tool

Each server entry accepts tools.include and tools.exclude. This entry registers the posthog__read tool and nothing else from the connector, which also hides the tools of every other platform in the workspace. It hides tools from Hermes Agent only. For a hard read-only guarantee, give the PostHog key read scopes alone.

mcp_servers: ryze: url: "https://connector.get-ryze.ai/mcp" auth: oauth tools: include: - posthog__read

A remote or headless host

When Hermes Agent runs on a server and your browser is on another machine, the local callback cannot reach the browser. The Hermes Agent documentation lists these ways around it: paste the redirect URL back at the prompt, forward the callback port with ssh -N -L, or complete the sign-in from the Desktop app or dashboard. A device-code flow (hermes mcp login ryze --flow device) works only when the authorization server advertises a device endpoint, and this guide did not check whether the Ryze AI connector does.

hermes mcp login ryze

The messaging gateway and unattended runs

A running gateway never opens a browser, so sign in with hermes mcp login ryze first. The documentation says the gateway watches config.yaml and connects a newly added entry within about a minute. A server whose first connect failed because no token was cached waits for a login.

Reloading and timeouts

The Hermes Agent pages read for this guide give no default for the length of a tool call. Hermes Agent waits for the browser sign-in for 5 minutes, and oauth.timeout on the entry makes it wait longer. connect_timeout on the entry bounds the first connection. PostHog stops a query after 10 seconds of execution time, so a slow HogQL query comes back as an error rather than a late answer.

mcp_servers: ryze: url: "https://connector.get-ryze.ai/mcp" auth: oauth oauth: timeout: 600 connect_timeout: 60

Hermes Agent and MCP: the facts in one table

TopicWhat the Hermes Agent documentation says
Where servers are added~/.hermes/config.yaml under mcp_servers, or with hermes mcp install for the catalog of Nous-approved servers. The documentation's catalog examples are deepwiki, linear and github, and the page names no PostHog entry
TransportsLocal stdio servers (command and args) and remote HTTP servers (url)
Sign-inauth: oauth does the browser sign-in. Hermes Agent identifies itself with a Client ID Metadata Document on servers that support one and falls back to dynamic client registration on those that do not. Static bearer headers also work
Token storage~/.hermes/mcp-tokens/<server>.json, written with 0600 permissions and reused until refresh fails
Tool namesmcp_<server_name>_<tool_name>, so a server named ryze registers its tools under names that start with mcp_ryze_
ApprovalA server entry can set trust: untrusted so that write-capable calls ask first. Tools annotated readOnlyHint stay unprompted. /yolo, --yolo and approvals.mode: off skip that prompt
Filteringtools.include and tools.exclude per server, with glob patterns. enabled: false skips a server. hermes mcp configure opens a checklist of the tools
Reload/reload-mcp, or a new session. Editing config.yaml inside a running CLI session reloads MCP with a 30-second timeout
ProxiesHTTP and SSE servers honor HTTPS_PROXY, HTTP_PROXY, ALL_PROXY and NO_PROXY
PlanNone. Open source, self-hosted

Sources: Hermes Agent MCP documentation, NousResearch/hermes-agent on GitHub. Opened November 19, 2026.

Other clients

The same URL works in other MCP clients. PostHog is connected once in Ryze AI, so nothing changes on the PostHog side.

Claude Code

Run claude mcp add --transport http ryze https://connector.get-ryze.ai/mcp, then sign in from /mcp. Add --scope user to make the server available in every project. PostHog is connected once in Ryze AI, so the PostHog side does not change from client to client.

Claude and ChatGPT

In Claude open Customize > Connectors, choose Add custom connector and enter https://connector.get-ryze.ai/mcp. In ChatGPT the full custom MCP flow is for Business and Enterprise/Edu plans, and Pro users get read and fetch permissions only. If the connector was added before PostHog was connected, remove and re-add it, then start a new chat.

Codex and Cursor

Codex takes codex mcp add ryze --url https://connector.get-ryze.ai/mcp. Cursor reads the same URL from its mcp.json file. The Ryze AI developers page has the matching config for each.

PostHog key scopes and Hermes settings

Two sign-ins are involved: a PostHog personal API key for PostHog access and Ryze AI for the connector. Three things decide what Hermes Agent can reach.

Which project
One PostHog project per connection. The key can cover several projects, but you pick one Project after connecting. Another project can live in another Ryze AI workspace.
What Hermes Agent can change
The scopes on the personal API key. A key with read scopes only keeps Hermes Agent read-only, because PostHog rejects any change. Add write scopes when you want it to edit objects.
What Hermes Agent asks first
The trust, tools.include and tools.exclude settings on the ryze entry decide which tools Hermes Agent sees and when it asks first. They are client-side settings and do not change what the PostHog key may do.

The personal API key and its scopes

The connection runs on the personal API key you saved in Ryze AI. PostHog treats that key as the user who created it, narrowed to the scopes and projects set on the key. Most scopes have a read level and a write level, and the write level includes read. The middle column follows PostHog's description of each scope.

ScopeWhat PostHog says it coversIn the key Ryze AI asks for
query:readSQL and insight queries on the project's data, plus lookups of events, sessions and web vitals. It has no write levelYes. This is the Query scope, and every HogQL question needs it
insight:read, insight:writeInsights with their queries, results and explanations, and anomaly rates for trends insightsRead is the Insight scope. Add write before Hermes Agent creates or edits an insight
dashboard:read, dashboard:writeDashboards with their tiles, text cards, widgets and layout, and running the insights on themNot named by Ryze AI. Add read if a dashboard request comes back with a 403 that names it
feature_flag:read, feature_flag:writeFlags with their release conditions, variants, payloads and rollout historyNot named. Write is needed before Hermes Agent can create or change a flag
cohort:read, cohort:writeCohorts with their filters, static member lists, calculation history and where they are usedNot named. Creating or deleting a cohort needs write
person:read, person:writePersons and their properties, including updating, splitting and deleting persons and their dataNot named
event_definition:read, event_definition:writeEvent definitions, schemas and property groups, with tags and generated type definitionsNot named. Add read if listing event definitions returns a 403

Not part of the default connection: any write scope, a second project in the same workspace, and the MCP Server key preset, which PostHog made for its own MCP server. Ryze AI's description of its PostHog tools does not mention session recordings, experiments, surveys or error tracking, so this guide does not claim Hermes Agent reaches them through Ryze AI.

Some PostHog rules about keys matter here. PostHog shows the key value once, when you create it, and one user can hold up to 10 keys. A key can reach the whole organization, chosen projects or every project, and its scopes can be changed later, so a missing read scope does not call for a new key. If GitHub secret scanning finds an exposed phx_ key, PostHog rolls it and emails its owner.

Sources: PostHog API scopes, personal API keys. Checked November 14, 2026. Which scopes Ryze AI asks for comes from the Ryze AI PostHog facts, not from PostHog.

Removing access

To remove access, click View on the PostHog card in Ryze AI and disconnect with the trash icon, or delete the personal API key in PostHog. In Hermes Agent, set enabled: false on the ryze entry or delete it from config.yaml, and delete ~/.hermes/mcp-tokens/ryze.json to drop the cached tokens.

What Hermes Agent can do in PostHog

The Ryze AI connector exposes 3 PostHog tools: 1 read, 1 write and 1 delete. Each works on the project you picked. In Hermes Agent the names appear with the mcp_ryze_ prefix.

ToolTypeWhat it does
posthog__readReadReads PostHog data for the connected project (projects, insights, dashboards, feature flags, cohorts, persons, event definitions) and runs HogQL analytics queries.
posthog__writeWriteCreates or updates PostHog objects such as insights, dashboards, feature flags and cohorts.
posthog__deleteDeleteDeletes a PostHog object by path, for example a cohort. The deletion is irreversible.

Approvals in Hermes Agent

In Hermes Agent, approval for MCP tools is set per server. Adding trust: untrusted to the ryze entry makes every write-capable call on that server ask first, and tools annotated readOnlyHint stay unprompted. The documentation names no default for an entry without a trust key, so set it explicitly if you want the prompt. /yolo, --yolo and approvals.mode: off skip the prompt. Whether the Ryze AI tools carry the readOnlyHint annotation was not checked, so expect prompts on some reads as well. Separately, the Ryze AI chat has its own Tool approvals picker that starts on Skip all approvals, and scheduled tasks run without an approval step. A key with read scopes only keeps any client read-only.

mcp_servers: ryze: url: "https://connector.get-ryze.ai/mcp" auth: oauth trust: untrusted

PostHog limits in numbers

The first row is the Ryze AI rule for queries. The rest are PostHog's published figures in PostHog's own units, last row included. None was measured for this guide. Most of PostHog's limits count per personal API key, and the key behind your Hermes Agent chats is the one you saved in Ryze AI.

LimitNumber
HogQL through Ryze AIA time window and 1,000 rows at most in each query. This bound comes from Ryze AI, not from a PostHog page
Analytics endpoints240 calls a minute and 1,200 an hour per personal API key, for work such as calculating insights and retrieving persons
Query endpoint2,400 an hour. PostHog's queries page sets it per project: 240 a minute, 3 queries running at once, 60 threads and 10 seconds of execution per query, and up to 30 seconds in the queue. Some older accounts stay on 120 queries an hour
Hourly read budgetQueries sent with a personal API key spend a per-project budget counted in bytes read. PostHog publishes no figure for it. When it runs out the answer is a 429 with api_queries_budget_exceeded, and cached results keep working
Other reads and changes480 calls a minute and 4,800 an hour per personal API key on the remaining create, read, update and delete endpoints
Rows in one API query100 by default, and up to 50,000 with your own LIMIT. OFFSET gets HTTP 400 with a personal API key, so PostHog points to paging on timestamp
Whose budget it isMost limits count per personal API key rather than per team, and PostHog says it does not offer higher ones
PostHog's own MCP server480 requests a minute and 4,800 an hour per credential, and opening a session or listing tools counts as a request. This applies to PostHog's server, not to the Ryze AI connector

Sources: PostHog API overview, queries, personal API keys, MCP FAQ. Checked November 14, 2026. The first row is a Ryze AI rule from the Ryze AI PostHog facts.

Tested calls

None. No PostHog call was run through the Ryze AI connector for this guide, so this page has no table of tested calls and quotes no output. The tool names and what they do come from Ryze AI's own record of the connector.

First 3 prompts to run

Each answer comes from your own PostHog project, so this page describes what comes back instead of showing sample numbers. Hermes Agent picks the tool itself; the names are listed so you can check which one ran.

1. Review what exists in the project

List the dashboards, feature flags and cohorts in my PostHog project, with each name and when it was last changed if that is available.

Tools used: posthog__read

What comes back: Lists of dashboards, flags and cohorts as plain text in Hermes Agent. Nothing in PostHog is changed. These lists may need the dashboard, flag and cohort read scopes, which the Ryze AI setup steps do not name. If PostHog answers with a 403 that names one, add that scope to the key in PostHog.

2. Run a bounded HogQL query

Run a HogQL query on pageview events for the last 14 days, grouped by day, with a limit of 100 rows, and show the result as a table.

Tools used: posthog__read

What comes back: A table of daily pageview counts from your own project. Ryze AI expects every HogQL query to carry a time window and no more than 1,000 rows, and this prompt sets both.

3. Draft a feature flag

Create a feature flag named new-onboarding-test that is off for everyone, and tell me its key and the URL path to open it.

Tools used: posthog__write

What comes back: A new flag in your project, if the API key has the feature_flag:write scope. With a read-only key, PostHog rejects the change and Hermes Agent reports the error.

When you want a change, describe it in plain words. With trust: untrusted on the ryze entry, Hermes Agent asks before it runs a write-capable call. Replace the names in the prompts with ones from your project.

10 jobs to hand to Hermes Agent

Ten jobs you can give Hermes Agent, each beside the tool that does the work and the PostHog scope it relies on. None was run for this guide. The last column assumes the key from the setup steps: read access to the project, the Query and Insight scopes, and nothing more.

JobAsk Hermes AgentToolScopeWorks on the default key
Daily pageviewsRun a HogQL query that counts pageviews per day over the last 7 days, 1,000 rows at most.posthog__readquery:readYes
Traffic by UTM sourceUse a HogQL query over sessions to split last month's sessions by UTM source.posthog__readquery:readYes
Signup funnel stepsRead my saved signup funnel insight and give the number of people at each step.posthog__readinsight:read, query:readYes
Dashboard namesName every dashboard in the project so I can point to one later.posthog__readdashboard:readOnly if the key has dashboard:read. PostHog's rule for a missing scope is a 403 that names it
Flags before a releaseShow each feature flag with its current rollout settings.posthog__readfeature_flag:readOnly if the key has feature_flag:read
Cohort definitionsList the cohorts and the conditions behind each one.posthog__readcohort:readOnly if the key has cohort:read
Which events existList the event definitions before we plan a report.posthog__readevent_definition:readOnly if the key has event_definition:read
Campaign dashboardMake a trends insight for landing page signups and place it on a new campaign dashboard.posthog__writeinsight:write, dashboard:writeNo. The key needs both write scopes
Campaign cohortCreate a cohort of the people who came from one campaign.posthog__writecohort:writeNo. The key needs cohort:write
Remove a test cohortDelete the test cohort I name here. I know this cannot be undone.posthog__deletecohort:writeNo. It needs cohort:write, and trust: untrusted makes Hermes Agent ask before the call

7 ways to connect Hermes Agent to PostHog

Seven ways to connect Hermes Agent to PostHog. This guide treats the Ryze AI connector as the main way: the PostHog key stays in Ryze AI, and Hermes Agent holds one address and a browser sign-in. Each setup cell says how the route would work in Hermes Agent, which accepts any remote HTTP MCP server by url. PostHog's pages name no Hermes Agent setup, and none of these routes was run in Hermes Agent.

WayRuns whereSetup in Hermes AgentWhat it reaches in PostHogPrice
Ryze AI connector (the main way)Hosted by Ryze AI, one address for every userA ryze entry with url and auth: oauth, then hermes mcp login ryze. Or a bearer header with a Ryze AI API key read from ~/.hermes/.env3 tools on one project: insights, dashboards, flags, cohorts, persons, event definitions and HogQL. Changes need write scopes on the keyPaid Ads is $89 per month with a 7-day free trial
PostHog MCP serverHosted by PostHog at https://mcp.posthog.com/mcp. Sign-in sends you to your US or EU regionA url entry named posthog under mcp_servers. PostHog recommends OAuth. Where the client has none, PostHog says to create a personal API key with the MCP Server preset and send it as an Authorization: Bearer header, which maps to the headers key. Whether PostHog's OAuth works with Hermes Agent's registration methods was not checkedProduct analytics, session replay, feature flags, experiments, error tracking, logs, surveys, SQL, data pipeline destinations and more: 1,284 tools in 78 categories, to read and write. ?readonly=true drops the tools that change data and ?features= narrows the setFree to connect and call. Some tools use LLMs and may be billed as PostHog AI spend
ComposioHostedComposio lists Hermes among the assistants on its PostHog toolkit page. A url entry for https://connect.composio.dev/mcp, then Authenticate and Allow access. PostHog is linked inside Composio with your own API key686 tools and 11 triggers in its Posthog toolkitThe free plan stops at 10,000 app tool calls a month. Standard is $18 per seat per month
Zapier MCPHostedAdd Zapier's MCP URL as a remote HTTP server and sign in. Zapier's page does not list Hermes Agent among the clients recorded for this guideOne action, Capture Event. No reads are listedEach successful tool call uses two tasks from your Zapier plan
AnythingMCPAnythingMCP Cloud in Frankfurt, or your own Docker installSet POSTHOG_HOST and POSTHOG_PERSONAL_API_KEY with a phx_ key, then add your AnythingMCP server URL as a url entry and sign in with OAuth. A phc_ project token does not work. Hermes Agent is not named on its guide28 tools, such as posthog_run_hogql_query. Its three delete tools are off at install, and it does not create or edit dashboards, cohorts or experimentsCloud from 19 euros per month with a 7-day free trial. The self-hosted Community edition is free for up to 3 active users
Coupler.ioHosted data export serviceEnter a PostHog private API key and your instance domain, pick a data entity and a start date, then choose Custom MCP as the AI destination. Hermes Agent is not named on its PostHog page7 entities collected on a schedule: projects, cohorts, feature flags, persons, events, annotations and insightsA free plan with manual refresh. Starter is $32 per month, or $24 billed annually
Windsor.aiHosted. PostHog is a data source and Windsor MCP a destinationConnect PostHog as a data source at onboard.windsor.ai, then add Windsor MCP in the client. Windsor's pages name ChatGPT, Claude, Perplexity, Cursor, Grok, Gemini and others, not Hermes Agent, and its setup steps are plugin installs101 PostHog fields, 6 metrics and 95 dimensions, in reports such as insights, cohorts, events and feature flagsA free plan: a 30-day trial, then 1 user, 1 data source and 1 account. Basic is $23 per month, or $19 billed annually. Every plan includes Windsor MCP

Ryze AI's older post, How to connect Google and Meta Ads with Hermes Agent, sets up Composio for ad platforms. It is a different route and several servers can share config.yaml under different names. A ryze entry and a posthog entry can sit side by side. With the Ryze AI connector your PostHog data passes through Ryze AI.

PostHog options that do not fit Hermes Agent

  • PostHog's AI plugin, Claude connector and ChatGPT plugin. They sit on the same hosted server, but the plugin listings and the Claude and ChatGPT pages name those assistants, not Hermes Agent. In Hermes Agent, add the server address yourself as in the table.
  • Enterprise-managed authorization. PostHog documents admin-managed sign-in for Claude alone, on a Claude Team or Enterprise plan with PostHog's Scale or Enterprise add-on. Hermes Agent is not named.
  • PostHog CLI (posthog-cli). It runs PostHog API calls and SQL queries from a shell. PostHog's CLI page names Claude Code, Codex and another coding agent, and does not name Hermes Agent, so this guide gives no steps for it.
  • Retired or unrelated PostHog packages. The standalone PostHog/mcp repository is archived, with the server moved into PostHog's main repository, and @posthog/agent-toolkit is deprecated in favor of the MCP server. The npm package @posthog/mcp is an SDK for tracking tool use in your own MCP server and does not connect Hermes Agent to PostHog data.

Which one to pick

Ryze AI is the best way to connect Hermes Agent to PostHog when you want one project's product analytics next to your ads and SEO tools behind a single address, with one caveat: it reaches less of PostHog than PostHog's own server, through 3 tools on one project per connection, and it needs a Ryze AI plan or trial. If PostHog is the only platform you need in Hermes Agent, PostHog's own server is the simpler pick. It is free and also reaches session replay, experiments, error tracking, surveys and logs, though PostHog documents no Hermes Agent setup for it and 1,284 tools is a long list to register, so add ?features= or ?tools= to the address. Composio and Zapier make sense if your team already runs them, though Zapier only captures events. AnythingMCP suits a team that wants to host the server itself, and Coupler.io fits when PostHog entities collected on a schedule are enough.

Sources: PostHog MCP server docs, MCP FAQ, posthog.com/mcp, Composio PostHog toolkit, Composio pricing, Zapier MCP for PostHog, Windsor.ai PostHog fields, Coupler.io PostHog source, AnythingMCP on GitHub, Hermes Agent MCP documentation. Checked November 14, 2026.

Hermes Agent, PostHog and MCP: dates that matter

DateWhat happened
February 2024PostHog personal API keys made before this month use legacy hashing and carry a Legacy tag. Rolling a key upgrades it.
May 4, 2026PostHog's MCP server gained an explicit mode choice through the x-posthog-mcp-mode header or the mode query parameter.
August 3 to October 15, 2026Nine changelog entries added new tools to PostHog's MCP server, between 5 and 44 at a time.
October 1, 2026PostHog stopped asking MCP clients with non-rotating refresh tokens to authorize again after roughly 67 days.
November 14, 2026PostHog's MCP page counted 1,284 tools in 78 categories.
November 19, 2026The Hermes Agent GitHub releases page listed canary pre-releases of v0.23.2, the newest published that day. Its MCP documentation can change between builds, so run hermes mcp test ryze after an update.

Sources: Hermes Agent releases, personal API keys, PostHog changelog, posthog.com/mcp, PostHog/ai-plugin, @posthog/wizard on npm. Checked November 14, 2026.

How this guide was tested

The Ryze AI connector facts were recorded on November 8, 2026, the PostHog pages behind this guide were read on November 14, 2026, and the Hermes Agent documentation was read on November 19, 2026. Each row is something that was read, with what it says.

What was runWhat came back
The Ryze AI record of the PostHog connection3 tools, posthog__read, posthog__write and posthog__delete, on one project per connection, with a phx_ key and a host. The tool list was not loaded from a signed-in client.
PostHog pages on API keys, scopes, queries and its MCP server, read on November 14, 2026The scope names, the limits and the own-server details on this page. No PostHog call was run.
The Hermes Agent MCP documentation, read on November 19, 2026It describes remote HTTP servers with auth: oauth, the Client ID Metadata Document and dynamic client registration, the token cache path, hermes mcp login, /reload-mcp, tool filters and the trust setting. Those pages are the source of every Hermes Agent step on this page.

Not run for this guide: any call to PostHog and every step inside Hermes Agent. No PostHog test account was connected, so the page shows no PostHog output and no table of tested calls. Hermes Agent was not installed, so hermes mcp login ryze and the config.yaml entries follow its documentation as read on November 19, 2026. On paper the sign-in fits: Hermes Agent falls back to dynamic client registration, which is the only registration method the Ryze AI connector supports. The Ryze AI help center has no page for Hermes Agent, and Ryze AI was not tested there.

Hermes Agent and PostHog errors and fixes

No PostHog call was run for this guide, so this section quotes no message from a test. The messages in the table come from PostHog's documentation, and the fixes follow Ryze AI's, PostHog's and Hermes Agent's documentation.

Messages PostHog documents

These responses are documented by PostHog. They are not results of a test for this guide.

MessageWhat it meansWhat to do
Invalid Personal API key. (401, invalid_personal_api_key)PostHog says the personal API key used for authentication is invalid.Create a new key in PostHog, then disconnect and reconnect PostHog in Ryze AI with it.
403 that names a missing scopeThe request needs a scope the key does not have.Add that scope to the key in PostHog. Scopes can be changed after the key is made.
api_queries_budget_exceeded (429 with 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 time in the Retry-After header, then ask again.
HTTP 400 on OFFSETPostHog rejects OFFSET paging on the query endpoint for personal API keys.Ask Hermes Agent to page on a timestamp column instead.

Other problems

The tools do not show up in Hermes Agent
Start a new session or run /reload-mcp. Then run hermes mcp test ryze, which reports what the server answered. Check that no tools.include or tools.exclude filter hides them and that the entry is not set to enabled: false. If PostHog was connected in Ryze AI after the login, run the login again and start a new session.
The browser sign-in times out
Editing config.yaml inside a running session triggers an automatic MCP reload with a 30-second timeout, too short for OAuth. Add the entry, then run hermes mcp login ryze from a fresh terminal. If you need longer than 5 minutes to approve, set oauth.timeout on the entry.
Hermes Agent cannot reach the callback on a server
Paste the redirect URL back at the prompt, forward the callback port over SSH, or sign in from the Desktop app or dashboard.
A 401 or 403 from the connector
Run hermes mcp login ryze again. With an API key, check that RYZE_API_KEY is set in ~/.hermes/.env and that the header reads Bearer followed by the rk_live_ key.
PostHog rejects a change
The personal API key has read scopes only. Add the write scopes you need to the key in PostHog, or create a key with them and reconnect PostHog in Ryze AI.
A query is refused or returns nothing
Ryze AI expects a time window and a row limit of 1,000 at most on every HogQL query. Ask for both in the prompt.
The wrong project or region answers
One project per connection, and the host decides the region: https://us.posthog.com, https://eu.posthog.com or your self-hosted URL. Click View on the PostHog card, disconnect with the trash icon and connect again with the right host and Project.
The one-time connection link is refused
PostHog holds one connection per workspace, and the link is refused while it is already connected. Switch projects or keys in the Ryze AI app.
The tools reply that a subscription is required
The PostHog tools need an active Ryze AI plan or free trial. Start the trial, then retry.
Requests fail behind a proxy
HTTP servers honor HTTPS_PROXY, HTTP_PROXY and NO_PROXY, so set them in the environment Hermes Agent starts from.

Terms on this page

Hermes Agent
An open source AI agent from Nous Research that you run on your own machine or server. Its MCP servers are listed in ~/.hermes/config.yaml under mcp_servers.
MCP
The Model Context Protocol, an open standard that lets large language models call tools on outside servers. The Ryze AI connector is a remote MCP server.
Personal API key
A PostHog key beginning with phx_ that acts as the user who made it, within the scopes and projects set on it.
Scope
One area of PostHog a key may touch, such as insights or feature flags, at a read level or a write level.
Host
Where your PostHog account lives: https://us.posthog.com, https://eu.posthog.com, or your own domain when self-hosted.
HogQL
PostHog's SQL, a wrapper around ClickHouse SQL. PostHog uses the name for SQL access internally.
Insight
The main building block of product analytics in PostHog. Trends, funnels and retention are types of insight.

Sources: PostHog SQL, insights, personal API keys, Hermes Agent MCP documentation. Checked November 14, 2026.

Frequently asked questions

How do I connect Hermes Agent to PostHog?

Connect PostHog in Ryze AI under Integrations with a phx_ personal API key and your host, pick the Project, then add a ryze entry with the URL https://connector.get-ryze.ai/mcp and auth: oauth under mcp_servers in ~/.hermes/config.yaml. Run hermes mcp login ryze, click Allow connection, then start a new session or run /reload-mcp.

Is there a PostHog MCP connector for Hermes Agent?

Yes, the Ryze AI connector at https://connector.get-ryze.ai/mcp. It is a remote MCP server with 3 PostHog tools, and Hermes Agent supports remote HTTP servers with OAuth. PostHog also runs its own hosted server at https://mcp.posthog.com/mcp, though its pages name no Hermes Agent setup.

Which should I use, Ryze AI or PostHog's own server?

The Ryze AI connector puts PostHog into Hermes Agent next to the other platforms on the same URL, but it has 3 PostHog tools and one project per connection. PostHog's own server is free and covers more of PostHog, such as error tracking and experiments, but its setup for Hermes Agent was not checked.

Does Hermes Agent have a PostHog MCP server built in?

The Hermes Agent documentation page read for this guide names no PostHog entry. Its catalog examples are deepwiki, linear and github. Add PostHog through a url entry in config.yaml, such as the Ryze AI connector.

Which PostHog API key do I need?

A personal API key starting with phx_, with read access to the projects you want and the Query and Insight scopes. Add write scopes only if Hermes Agent should change objects. Also enter your host: US Cloud, EU Cloud or a self-hosted URL.

Can Hermes Agent change my PostHog project?

Yes, within the key's scopes. Through Ryze AI it can create or update insights, dashboards, feature flags and cohorts, and delete those objects when you ask. A read-only key keeps it read-only, because PostHog rejects any change. To have Hermes Agent ask first, set trust: untrusted on the ryze entry.

How do I make it read-only?

Give the PostHog key read scopes only, which is a hard guarantee. In Hermes Agent you can also list posthog__read under tools.include on the ryze entry, which hides the write and delete tools from the agent but is not enforced by PostHog.

Does Hermes Agent need a paid plan for MCP?

No. Hermes Agent is open source under the MIT license per its GitHub repository and self-hosted. The Ryze AI side needs a plan or trial for the PostHog tools: Paid Ads at $89 per month with a 7-day free trial.

Should I use OAuth or an API key?

Use OAuth when you can open a browser, because hermes mcp login ryze handles the sign-in and Hermes Agent refreshes the tokens. Use a Ryze AI API key in a bearer header for servers and scripts where nobody can click Allow connection. The Ryze AI key is separate from the phx_ PostHog key.

Does it work on a headless server or a VPS?

Yes, with an extra step for the sign-in. Paste the redirect URL back at the prompt, forward the callback port over SSH, or use the Desktop app or dashboard. An API key avoids the browser entirely.

Can I connect more than one PostHog project?

Each connection holds one project. To switch, click View on the PostHog card in Ryze AI, disconnect with the trash icon and connect again. Another project can live in another Ryze AI workspace.

Does it work with self-hosted PostHog?

Yes. When you connect PostHog in Ryze AI, enter your self-hosted URL as the Host instead of the US Cloud or EU Cloud address.

Why do the PostHog tools not show up in Hermes Agent?

Start a new session or run /reload-mcp, then run hermes mcp test ryze. Check that no tools.include or tools.exclude filter hides them and that the entry is not set to enabled: false. Also confirm that PostHog is connected in Ryze AI before the login.

How much does it cost?

Adding the connector is free. The PostHog tools need a Ryze AI plan or trial: Paid Ads is $89 per month with a 7-day free trial, SEO Starter is $129 per month with a 3-day free trial, Agency is $279 per month, and annual billing takes 30% off the first year. Hermes Agent itself is free software.

How do I disconnect PostHog from Hermes Agent?

Set enabled: false on the ryze entry or delete it from config.yaml, and delete ~/.hermes/mcp-tokens/ryze.json. To remove PostHog access itself, disconnect PostHog on its card in Ryze AI and delete the personal API key in PostHog.

Sources

Checked November 14, 2026.

Connect PostHog to Hermes Agent

One Ryze AI connection gives Hermes Agent 3 PostHog tools. Paid Ads is $89 per month with a 7-day free trial; annual billing takes 30% off the first year.

PostHog MCP server

Need help? Read the Ryze AI PostHog guide or email Ryze AI support.