Claude + Statsig: What the Integration Can (and Can't) Do in 2026
There’s no Anthropic directory connector for Statsig — but Statsig ships its own official MCP server, so Claude Code can already read your gates, experiments, and metrics. The server is remote and vendor-hosted at https://api.statsig.com/v1/mcp, it authenticates over OAuth, and it exposes read tools for feature gates, A/B experiments, metrics, dynamic configs, layers, segments, and parameter stores. If you’d rather not use the hosted server, the Console API plus a community MCP server covers the same ground.
Here’s each route, when to pick it, and what to use when you want Statsig work that happens without you at the keyboard.
Route 1: Statsig’s official MCP server (the one to use first)
Statsig operates a remote MCP server at https://api.statsig.com/v1/mcp. It’s the recommended path because Statsig maintains it and because it authenticates over OAuth rather than a long-lived key — the client walks you through the login the first time you use it, and you get scoped access instead of pasting a full-power key into a config file.
Read-only access lets an agent explore your project; write-capable access (creating or deleting gates, managing parameter stores) requires an API key with write permissions. Read-only users can connect and use every read tool, which is what most Claude workflows want anyway.
To add it to Claude Code, register it as an HTTP MCP server:
claude mcp add --transport http statsig https://api.statsig.com/v1/mcpThen, in a session, Claude can:
- Read feature gates — pull the rollout rules and status of any gate.
- Read experiments — surface an A/B test’s setup, variants, and current state (2026 updates made experiment reads noticeably leaner).
- Read metrics — look up metric definitions and list what’s tracked in the project.
- Reason across dynamic configs, layers, segments, and audit logs — answer “which gates target our enterprise segment?” from live data.
Statsig documents the server for Cursor, Codex, and ChatGPT as well as Claude Code, so the same endpoint works across coding assistants. Note the ownership context: OpenAI acquired Statsig in September 2025 for $1.1B and made founder Vijaye Raji its CTO of Applications. Statsig continues operating independently out of Seattle, and the MCP server remains open to any MCP client, Claude included.
Route 2: Console API + a community MCP server
If you’d rather authenticate with a key than OAuth — or you want write access without OAuth scopes — you can point Claude at Statsig’s Console API through a community MCP server. GeLi2001/statsig-mcp on GitHub is an early, low-adoption project (a handful of stars) that wraps the Console API into roughly 27 tools spanning feature gates, experiments, dynamic configs, segments, and monitoring. It reads your Console API key from a STATSIG_CONSOLE_API_KEY environment variable.
Add it to Claude Desktop by editing claude_desktop_config.json:
{
"mcpServers": {
"statsig": {
"command": "npx",
"args": ["-y", "statsig-mcp", "--api-key", "YOUR_CONSOLE_API_KEY"]
}
}
}Because a Console API key grants full read/write access to the project, treat it like a production secret — never expose it client-side, and prefer a scoped key where you can. Because it’s community-maintained, verify the current package and tool list before you rely on it; the official server (Route 1) is the safer default.
Route 3: Let Claude write against the Console API directly
For one-off analysis or scripting, you don’t need any MCP server. The Console API is Statsig’s CRUD REST API for everything you can do in the console UI — gates, experiments, dynamic configs, metrics, and project settings. You create a Console API key at console.statsig.com/api_keys and send it in the STATSIG-API-KEY header:
GET https://statsigapi.net/console/v1/experiments/{id}— read an experiment.GET https://statsigapi.net/console/v1/metrics/list— list all metrics.GET https://statsigapi.net/console/v1/metrics/{id}— read a metric definition.
Ask Claude (in Claude Code, or with the code interpreter) to write a script that pulls an experiment’s status or diffs metric definitions, and you get repeatable Statsig automation you can run yourself. Mutation requests are rate-limited to roughly 100 per 10 seconds per project, so bulk changes need pacing.
The limits that actually matter
- No Anthropic directory connector. claude.ai doesn’t list Statsig as a first-party connector — the official MCP server is the supported path, and it’s aimed at coding clients like Claude Code, not the claude.ai web app.
- No triggers. Every route above runs inside a session you start. Nothing watches Statsig for a failing experiment or a gate flip and acts on it.
- Laptop-bound. A local MCP config or a script only runs while your machine is on and the client is open. Close the laptop and nothing happens.
- claude.ai on the web can’t reach a local/desktop MCP setup. The remote server needs a client that speaks MCP over HTTP; the community server and Console API scripts are local.
Claude + Statsig is excellent for asking questions about your experiments and gates in the moment — it’s not built for Statsig work that has to happen on a schedule or in response to an event.
If you want Statsig work that runs on its own: Carly
Claude answers when you ask. If you want experiment monitoring that fires on its own, that’s a different tool.
Carly reaches Statsig through your own Statsig Console API key (paste it on the integrations page) — one of 200+ native integrations plus anything with an API. Because Carly runs in the cloud on triggers and schedules, it does the parts a chat session can’t:
- When an experiment reaches significance, Carly emails you the readout — variant, metric lift, and a plain-English call on whether to ship.
- Every morning, Carly checks your live gates and flags anything rolled out to 100% for a week so stale flags don’t pile up.
- When a guardrail metric regresses, Carly posts the experiment name and delta to Slack before it becomes a launch-day surprise.
- No-code setup — you don’t write config. Carly interviews you about what you want watched and what should happen, in plain English, and builds the workflow.
- Actually sends — Carly drafts and sends through Gmail and Outlook, not just a draft you copy out.
- Connects to anything — 200+ native integrations plus any tool with an API via your own key, so a Statsig result can flow straight into Linear, a doc, or an email.
AI agents start at $35/month, and steps in a workflow that don’t use AI run free and unlimited.
Claude vs Carly
| Claude + Statsig | Carly | |
|---|---|---|
| Read/search gates & experiments | Yes (official MCP) | Yes (your API key) |
| Act in a session you start | Yes | Yes |
| Acts on triggers & schedules | No | Yes |
| Works while your laptop is closed | No | Yes |
| Sends email | No | Yes — Gmail + Outlook |
| Reaches other apps in one flow | One at a time | 200+ native + BYO-key |
| Pricing | Pro $20 / Max $100–$200 | AI agents from $35/mo |
Claude is the best way to interrogate your Statsig project on the spot; Carly is the way to make Statsig results reach you and other tools automatically.
Frequently Asked Questions
Does Claude work with Statsig?
Yes. Statsig ships an official remote MCP server at https://api.statsig.com/v1/mcp, and Claude Code can connect to it over OAuth to read gates, experiments, metrics, and more. There’s no claude.ai directory connector, so this is a coding-client path rather than a web-app one.
Is there an official Statsig MCP server in 2026?
Yes — Statsig operates and maintains it. It’s remote and vendor-hosted, authenticates over OAuth, and exposes read tools for gates, experiments, metrics, dynamic configs, layers, segments, and parameter stores; write tools need an API key with write permissions.
Can claude.ai on the web access Statsig?
Not directly. The official MCP server targets clients like Claude Code, Cursor, and Codex, and community servers run locally. The claude.ai web app has no Statsig connector in the directory.
Can Claude watch an experiment and alert me when it hits significance?
No — every Claude route runs inside a session you start, so nothing monitors Statsig on its own. For trigger-based alerts on experiment results or gate changes, use an always-on agent like Carly.
More: Carly integrations · How Claude connectors work · Claude + Linear · Claude + Slack · Claude vs Carly
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
