StatsigMCP Server

Statsig is a product-development platform combining feature management, experimentation, product analytics and session replay. Its official MCP server hands an agent 80+ tools over that data, which turns questions that normally mean clicking through a console into something you ask in the editor.

The coverage is broad. Experiments support creation, detail retrieval, results analysis with dimensional breakdowns, config updates and version history. Feature gates get full CRUD plus results analysis and the ability to kick off code cleanup once a flag is fully rolled out. Dynamic configs, layers, segments and parameter stores are all readable and writable, autotunes cover multi-armed-bandit experiments, and there are tools for metric definitions and metric sources, audit-log filtering, review workflows for gates and experiments, and querying raw events and clustered log patterns.

Authorisation follows the key you connect with: read-only users get every read tool; write operations require a Console API key with write permissions, so an agent can be given analysis access without the ability to flip a flag in production.

One detail worth knowing before you wire it up: Get_Experiment_Details_by_ID takes optional field filters that Statsig documents as cutting context consumption by 95–99%. Experiment payloads are large, and an agent that pulls them unfiltered will burn its window on metadata it does not need.

The server is remote — https://api.statsig.com/v1/mcp over streamable HTTP with a statsig-api-key header — and Statsig publishes setup guides for Claude Code, Cursor, Codex (desktop, CLI and IDE), the ChatGPT connector, and manual configuration for anything else that speaks MCP.