Best Claude MCP for Meta Ads — 2026 Rankings

Screenshot — Ryze AI MCP Meta audit output rendered inline as a Claude artifact in Claude Desktop.

Ryze AI was designed for Claude from day one, with deep Meta-specific tooling. The MCP exposes 42 tools with names like audit_creative_fatigue, find_high_frequency_ad_sets, draft_fresh_hook — verb-first names that match Claude’s tool-selection patterns. Responses come back as Markdown tables and JSON shapes Claude renders directly as artifacts, so a query about top-fatigued ad sets shows up as an inline sortable table with frequency, CPM, and ROAS columns, not a paragraph of prose.

The MCP ships with 16 Meta-specific prompt templates that Claude picks up automatically — type “/audit_meta” and Claude reaches for the right Ryze tool stack. Native Conversions API support, Instagram-specific tools (Reels CPM, Stories analytics, content publishing), and Business Manager auto-discovery all included. Measured uptime over 30 days: 99.6%, with the only downtime being a 14-minute Meta rate-limit retry that completed automatically.

Screenshot — Pulselane workflow exposes ~50 Claude-callable Meta tools through pre-built flow steps.

Pulselane exposes about 50 Meta tools to Claude, sourced from the visual workflows you build. The Claude integration is genuinely good because each workflow becomes a single named tool from Claude’s perspective — one workflow named “daily_meta_audit” encapsulates a multi-step pipeline pulling ad-set data, scoring fatigue, and posting Slack alerts. That’s closer to how Claude wants to think than dozens of granular Meta Marketing API calls.

Where Pulselane loses points: artifact formatting is generic JSON, so Claude can render it but with more friction. Prompt templates are limited (you build your own). Uptime in our 30-day Meta test: 99.3%, mostly clean except an 8-hour outage when Meta’s Marketing API hiccupped and Pulselane’s retry logic stalled. Solid Claude MCP, but designed multi-LLM and multi-platform rather than Claude-and-Meta-first.

Screenshot — Loomstack’s connector dashboard exposes 80+ Meta + multi-platform tools to Claude.

Loomstack exposes 80+ tools to Claude across Meta, Google Ads, Slack, GitHub, and HubSpot — the broadest surface in our ranking. Multi-platform agencies love this because Claude can pull data from Meta, TikTok, and Google in a single conversation. Claude tool-selection accuracy holds up surprisingly well at this size, mostly because Loomstack groups tools by platform.

The Claude-specific weakness on Meta: tool naming is generic API-style (meta_ads.get_adsets) rather than verb-first, so Claude leans on them less proactively. Output is unformatted JSON. No bundled Meta prompt templates. Claude works fine here, but you can tell the MCP wasn’t designed Claude-and-Meta-first — it was designed multi-LLM, multi-platform, and Meta is one of many surfaces.

Screenshot — Pivix mads-mcp source: exposes raw Meta Marketing API graph queries to Claude with read-only access.

Pivix exposes 14 tools centered on a powerful single tool: execute_graph_query. That one tool gives Claude effectively infinite query flexibility because the Meta Marketing Graph is the full underlying API. Power users love it because Claude can compose its own queries against any Meta entity. Casual users find it harder — Claude has to write Graph syntax which costs round-trips, and the response shapes are deeply nested.

Output is raw Meta API JSON with no formatting. No prompt templates. Self-hosted reliability depended on whether you set up Meta access-token rotation correctly — we hit a 53-hour effective downtime when our Business System User token expired during a Marketing API version bump and the auto-refresh path didn’t handle the new error code. For Claude-first use, Pivix mads-mcp is the “raw protocol” option: maximum flexibility, minimum hand-holding.

Screenshot — Tasknest exposes ~28 Meta Ads tools to Claude as task-shaped wrappers.

Tasknest exposes about 28 Meta Ads tools to Claude through their no-code task abstraction. Tool naming is friendly (find_high_frequency_ad_sets) which Claude reaches for naturally, but the gateway layer adds 250-450ms per call — noticeable when Claude chains multiple tools across Facebook and Instagram in one response.

No bundled Meta prompt templates, but Tasknest’s template marketplace has user-contributed Claude prompts for Meta workflows (creative-fatigue alerts, Reels analysis). Output is generic JSON. 99.4% measured uptime with one notable 28-minute outage during a Meta token rotation. Claude works fine here, but the per-task pricing model means heavy Claude conversations across multiple Meta accounts rack up bills faster than dedicated MCPs.

The marlowe community fork of Pivix exposes the same 14 Meta tools but with slightly better error messages back to Claude. That matters because Claude reads the error and decides whether to retry, fall back, or surface the problem to the user. Pivix’s upstream Meta API errors are terse (“OAuthException 200”); marlowe’s explain the failure mode in plain English (“access token expired, requires User permission re-grant”), which Claude handles much more gracefully.

Otherwise the same trade-offs as Pivix: read-only, raw JSON output, no prompt templates, single-maintainer reliability. Notable issue: the fork lagged 11 days behind a Meta Marketing API breaking change in our 30-day test (Meta deprecated a frequency-cap field), which broke the MCP for Claude users until the maintainer pushed a patch. For a Claude-first deployment on Meta, you’d ideally fork-the-fork or contribute upstream.