Statsig MCP server: 18 V3 tools, with discovery for the rest

A feature-flag connector rebuilt around grouped tools and a discovery layer, its legacy V1 server, and the documentation server people confuse with both.

Checked 6 October 2026 · from Statsig's own MCP documentation, including the V3 tool reference and the V3 migration guide

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 Statsig MCP server?

It is Statsig's server for the Model Context Protocol. Statsig is an experimentation and feature-flag platform, so a connected agent reaches gates, experiments and the metrics behind them.

V3 groups the surface by resource. Five read tools cover gates, experiments, dynamic configs, metrics and logs. Three create tools and three update tools cover gates, experiments and dynamic configs.

Other objects go through discovery. Layers, segments, Parameter Stores, Autotunes, holdouts, prompts, tags, dashboards and reviews are discovered operations. A layer groups experiments so that no user is in more than one of them.

Statsig sits in CI and release, reaches production observability through its logs explorer, and reaches nothing in agent observability or cost.

For an agent, Statsig answers what is gated, what is being tested and who approved it. Discovered operations can also propose, approve and commit a gate change on the same connection. It has no application metrics or traces, so it pairs with a telemetry connector.

What can an AI agent do with Statsig?

Start with get_context, which returns the project, permissions and review settings. For an operation without a direct tool, call discover_tools with resolve, then run the returned operationHandle. Handles expire, so fetch a fresh one each time.

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. The figures below measure the V1 server's 67 documented tools, read on 24 August 2026. On Claude they came to 3,669 to 4,619 tokens.

V3 sends 18 default tools in place of 67. A V3 figure needs a live tools/list, and the endpoint requires a Console key for that.

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 67 V1 tool entries in the vendor's own documentation, read 24 August 2026. That was the only public inventory.

The server could not be run for a live tools/list. So no input schemas are available. These figures cover each tool's name and description only. They are a floor, not the figure.

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 quarter.

Statsig MCP server limits: where answers come back incomplete

Two facts set the limits below. The same connection can propose a change and approve it. A separate Statsig server shares the brand and reaches none of your data.

How to configure Statsig MCP for agents

1. 1

   Pick V3 for a new connection

   Statsig recommends api.statsig.com/v3/mcp. V1 stays supported for existing connections, and connecting both duplicates tools.

2. 2

   Decide whether an agent may approve

   Approve, reject and commit actions for reviews run through api_destructive, which needs a confirmation.

3. 3

   Know which server you are configuring

   The Docs MCP is read-only over public documentation; the product endpoint needs OAuth or a Console key and reaches your data.

4. 4

   Understand the update semantics

   Most V3 update tools replace the whole resource. Partial edits, such as adding a tag, are discovered operations.

5. 5

   Scope the Console key

   Statsig states that the authenticated server accesses product resources according to the account's permissions.

Paste this into your agent’s instructions

The same connector can propose a change and approve it, and most updates replace rather than patch. The response must disclose each behavior.

When you answer using Statsig MCP tools, state:
- The project and object the answer covers.
- Whether you used the Docs MCP or the authenticated product
  endpoint. They return different things.
- Each discovered operation you ran, and whether it used
  api_write or api_destructive.
- Whether you created, approved, committed or canceled a
  review, or replaced an object with an update tool.
If no objects match, say that none matched the filters
above. Do not claim 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 pulling results on a schedule. Statsig's Console API covers the same objects and is the direct route there.

The connector earns its place on the review workflow. Creating a gate review and finding its eligible reviewers from the editor is not something a fixed report does.

For flags without the experimentation surface, LaunchDarkly's connector is narrower and its tools are all described.

Connect it when an agent participates in flag and experiment decisions. Use the Console API for reporting.

Statsig MCP server setup

The server is hosted at api.statsig.com/v3/mcp, so nothing is installed. Statsig recommends OAuth, which uses a Personal Console API key. A Console API key header also works.

Statsig publishes configuration guides for Codex, Cursor, Claude Code, a ChatGPT connector and a manual setup route.

claude mcp add --transport http statsig \
  https://api.statsig.com/v3/mcp

{"mcpServers":{"statsig":{
  "url":"https://api.statsig.com/v3/mcp",
  "headers":{"STATSIG-API-KEY":"<your-console-key>"}}}}

[mcp_servers.statsig]
url = "https://api.statsig.com/v3/mcp"
headers = { "STATSIG-API-KEY" = "<your-console-key>" }

Every other client

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

{"mcpServers":{"statsig":{"url":"https://api.statsig.com/v3/mcp",
  "headers":{"STATSIG-API-KEY":"<your-console-key>"}}}}

{"mcpServers":{"statsig":{"url":"https://api.statsig.com/v3/mcp",
  "headers":{"STATSIG-API-KEY":"<your-console-key>"}}}}

{"servers":{"statsig":{"url":"https://api.statsig.com/v3/mcp",
  "headers":{"STATSIG-API-KEY":"<your-console-key>"}}}}

{"mcpServers":{"statsig":{"url":"https://api.statsig.com/v3/mcp",
  "headers":{"STATSIG-API-KEY":"<your-console-key>"}}}}

{"mcpServers":{"statsig":{"url":"https://api.statsig.com/v3/mcp",
  "headers":{"STATSIG-API-KEY":"<your-console-key>"}}}}

{"mcpServers":{"statsig-v1":{"url":"https://api.statsig.com/v1/mcp",
  "headers":{"STATSIG-API-KEY":"<your-console-key>"}}}}

# Statsig's Docs MCP, for public documentation only.
# Configure it separately from the product endpoint above.

Read-only access, permissions and security

• Reviews run through discovery

  Approve, reject, commit and cancel actions for gates, experiments, dynamic configs and autotunes are discovered operations on api_destructive.

• Approval and commit are reachable from the same connection

  One key can propose a change and approve it. Project review policies still apply, and approving a review leaves it uncommitted.

• The agent supplies the confirmation

  Update tools and api_destructive require a confirmation object. Review each change before the agent sends it.

• Most updates replace rather than patch

  V3 update tools replace the whole resource, so fields left out are lost.

• Deletes are discovered operations

  Deleting an experiment or a Parameter Store runs through api_destructive.

• Permissions follow the key

  Statsig documents that the authenticated server accesses resources according to the account's permissions.

Troubleshooting

The endpoint returns 401

It accepts an active Console key only, and says so in the error. Other Statsig key types are rejected.

An update wiped fields you did not touch

Most V3 update tools replace the resource. Read the object first and send it back complete, or discover a partial-update operation.

The agent answered from documentation, not your data

It may have used the Docs MCP, which is a separate read-only server over public documentation.

A review will not commit

Approving a review leaves it uncommitted, and commit is its own discovered operation. Check eligibility with the eligible-reviewers operation for that object type.

A V1 tool name stops working

V3 uses different names. Statsig's migration guide maps each V1 tool, such as Get_List_of_Gates, to its V3 replacement.

Every tool appears twice

V1 and V3 are both connected. Statsig advises disabling the version you are not testing.

Statsig MCP server: Reference

What engineers report

Statsig documents its tools in a V3 reference, a V1 reference and a migration guide, so the counts below are its own.

Tool inventory read from Statsig's V3 tool reference, V1 tool reference and migration guide on 6 October 2026. Unauthenticated tools/list calls to the V1 and V3 endpoints both returned 401 that day, with a message stating that the endpoint accepts an active Console key only.

Should you connect an agent to Statsig?

Yes. Best for v3 over OAuth, with a key scoped to what the agent should decide and every api_destructive call reviewed.

Strongest fit

Statsig teams who want an agent inside the gate and experiment review process, not just reading it.

Main advantage

A short default tool list, with a discovery layer that still reaches reviews, layers, segments and logs.

Main weakness

Less common work takes two calls through discovery, and two server versions plus a docs server share the brand.

Operational risk

Approve and commit sit behind the same key as create, and most updates replace whole objects.

Frequently asked questions