PostHog MCP server: 1,080 tools, and how to scope them

The largest documented tool list in this guide, the query parameter that makes it usable, and what the free hosted endpoint costs.

Checked 6 October 2026 · from PostHog's own MCP documentation and the GitHub record for its repositories

One of 91 guides to monitoring and developer tools. Each covers what an AI coding agent can get out of the tool through MCP, the protocol that lets the agent query the tool directly.

What is the PostHog MCP server?

It is PostHog's server for the Model Context Protocol. PostHog covers product analytics, feature flags, experiments, error tracking, session replay, surveys and a data warehouse, and the tool list follows all of it.

The categories give the shape better than the total. AI observability carries 95 tools, customer analytics and signals 73 each, replay vision 55, experiments 50 and workflows 35.

365 of the 1,080 names are shaped for writing. They contain create, update, delete, destroy, launch, stop, archive, merge or split, and 70 of those are deletions.

PostHog has one live MCP server and one archived repository:

The archived repository is still what a search for the package name finds. The living code is in the monorepo under services/mcp, and the endpoint above is what clients connect to.

It is strongest in production observability and reaches agent observability and AI cost through its own analytics products.

For an agent, PostHog is the widest single connector among these vendor guides, and the width is the problem rather than the point. It answers questions about users, errors, flags and model spend. Deciding which of the 73 categories a session needs is the work that makes it usable.

What can an AI agent do with PostHog?

Every row above is reachable only if its category is in the session's scope. The tools query parameter is what puts it there, and PostHog states that tool names are stable so a scope can be written once.

What the tool definitions cost in context

A client that loads every tool definition at the start of a session spends this much context before the first question. Claude Code defers definitions by default through tool search, and loads a tool's definition when the agent needs it. The method below gives the version and date measured.

A context window is the amount of text a model can hold at once. On 28 August 2026, the 899 documented tools cost 38,690 to 51,946 tokens on Claude for a client that loads every definition upfront. That is a quarter of a 200,000-token window. The reference now lists 1,080, so the full list costs more today. The largest surface captured from a live tools/list across these guides is Rootly, at 252.

The descriptions are what keep it from being worse. PostHog writes one terse line per tool, often three words, so 899 definitions cost about what a hundred verbose ones would. Scoping the session with the tools parameter is still the whole job.

Encodings: GPT-5.6, GPT-5.5 o200k_base; GPT-4, GPT-3.5 cl100k_base; Claude Haiku 4.5 to Opus 5 count_tokens. Counted from the 899 tool entries in the vendor's own documentation, read on 28 August 2026. The reference listed 1,080 tools on 6 October 2026.

The hosted server answers tools/list only after authenticating, so no input schemas were counted. These figures cover each tool's name and description only. They are a floor.

The Claude columns are measured through the API's count_tokens endpoint, with an empty schema per tool. So they cover the same content as the GPT columns. The range there is two tokenizer generations. Haiku 4.5 and Opus 4.6 sit at the low end. Opus 5 and Sonnet 5 sit at the high end. The gap between them is about a third.

Can PostHog watch your own AI agents?

PostHog counts 95 tools in its AI observability category alone, so this table asks which agent signals those actually cover. Read from PostHog's own tools reference on 6 October 2026.

PostHog MCP server limits: where answers come back incomplete

Every limit below follows from a documented surface more than three times the size of any other here.

How to configure PostHog MCP for agents

1. 1

   Scope the session

   The tools query parameter narrows a connection to a named subset, and with 1,080 documented names it is required.

2. 2

   Decide about AI data processing

   Some tools use models internally and are only available when AI data processing is enabled in the organization settings.

3. 3

   Expect model usage to be billed

   Connecting and calling tools is free, and PostHog states that the tools using models internally may be billed as PostHog AI spend.

4. 4

   Let the region follow the account

   PostHog's authentication server routes to the US or EU region based on the account signing in, so the endpoint is the same either way.

Paste this into your agent’s instructions

With 1,080 documented names, an agent that cannot find a capability may simply be outside the session's scope. The block makes that distinction visible.

When you answer using PostHog MCP tools, state:
- The project the answer covers.
- The tool you called. Many names differ by one word across
  73 categories and return different resources.
- Whether a capability was missing because the session's
  tools scope excludes it, rather than because it does not exist.
When a call returns nothing, state that no results were
returned for the project and tool above. Do not state that
the data does not exist.
Write your answer in ASD-STE100 Simplified Technical
English. Use short sentences with one idea in each.

Do you need the MCP server at all?

Not for one known query. PostHog has a documented API, and calling it directly costs nothing in context.

The connector earns its place on breadth an API does not organize. An agent can move from an error to the flag that caused it to the replay that shows it, inside one session.

For a narrower question, a narrower connector is usually the better answer. PostHog documents more tools than any other vendor here, and that width is a cost as much as a capability.

Connect a scoped session for cross-product investigation. Call the API for a single known query.

PostHog MCP server setup

The server is hosted and free to connect to. The fastest route is PostHog's own wizard, which writes the configuration for seven clients.

Sign-in happens in a browser, and PostHog's authentication server sends the session to the US or EU region based on the account.

npx @posthog/wizard mcp add

claude mcp add --transport http posthog \
  https://mcp.posthog.com/mcp

{"mcpServers":{"posthog":{
  "url":"https://mcp.posthog.com/mcp"}}}

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

{"mcpServers":{"posthog":{"url":"https://mcp.posthog.com/mcp"}}}

{"mcpServers":{"posthog":{"url":"https://mcp.posthog.com/mcp"}}}

{"servers":{"posthog":{"url":"https://mcp.posthog.com/mcp"}}}

{"mcpServers":{"posthog":{"url":"https://mcp.posthog.com/mcp"}}}

{"mcpServers":{"posthog":{
  "url":"https://mcp.posthog.com/mcp?tools=docs-search,feature-flag-get-all"}}}

Read-only access, permissions and security

• 365 of the 1,080 documented names write

  They contain a writing verb, and 70 of those are deletions.

• Deletion sits beside reading

  session-recording-delete and delete-feature-flag are in the same categories as the tools that read the same resources.

• Some tools call models

  PostHog states that certain tools use models internally, are billed as AI spend, and are available only when AI data processing is enabled for the organization.

• Scope is the main control

  The tools query parameter is the documented way to decide what a session can reach, since there is no toolset switch on the server side.

Troubleshooting

The client struggles with the tool list

1,080 tool names are documented. Use the tools query parameter to scope the session to the categories the task needs.

An AI-powered tool is unavailable

Those tools require AI data processing to be enabled in the organization settings. Until it is, they do not appear.

The bill went up after connecting

Connecting and calling tools is free, but tools that use models internally may be billed as PostHog AI spend.

A guide points at an archived repository

PostHog/mcp was archived on 19 January 2026 when the code moved into the PostHog monorepo. Connect to the hosted endpoint instead.

PostHog MCP server: Reference

What engineers report

PostHog's server is official, hosted, free, and configurable in seven clients by one wizard command. The counts below come from its tools reference.

“I have my instance running in EU, so there is currently no way for me to use it. Maybe we could use headers to set the region?”

“Could you please share a detailed list of the capabilities and supported commands for this MCP? I'd like to help document them, but I want to make sure I'm capturing all the key features and use cases.”

Tool inventory read from PostHog's own MCP tools reference on 6 October 2026, by counting the tool names it lists and the per-category totals it prints. The hosted endpoint answers tools/list only after authenticating, so the figures come from documentation rather than from the server itself. Repository facts from the GitHub API.

Should you connect an agent to PostHog?

Yes. Best for a scoped session per task, using the tools query parameter on names PostHog documents as stable.

Strongest fit

PostHog teams whose questions cross products, and who are willing to write a tool scope for each kind of session.

Main advantage

The widest documented surface in this guide, free to connect, with a wizard that configures seven clients.

Main weakness

1,080 documented tools cannot be loaded at once, so a working configuration needs a scope.

Operational risk

365 tools that create, update or delete, including 70 deletions, sitting in the same categories as the tools that read.

Frequently asked questions