Documentation
MCP server (AI tools)
Splitlab runs a Model Context Protocol (MCP) server. Connect it to an AI tool and you can ask questions about your workspace in plain language: which test is winning, how revenue per visitor moved this week, or whether attribution is healthy.
| Server URL | https://mcp.splitlab.io/mcp |
| Transport | Streamable HTTP |
| Sign-in | OAuth 2.1 (most clients), or a workspace API key sent as a Bearer header |
| Access | Read-only, one workspace per connection |
Availability
The MCP server is rolling out gradually. If you don't see Connect AI tools under Workspace settings → API keys, contact Splitlab support to have it switched on.
What it can and cannot do
- Read-only. No tool creates, changes, pauses or deletes anything. An AI tool can't start a test, edit a variant or change traffic.
- One workspace. When you sign in, you choose the workspace the connection can read. It never reaches your other workspaces. Connect again to use a different one.
- No lead or payment data. Individual leads, contact details and billing are not exposed.
Tools
| Tool | What it returns |
|---|---|
whoami | The signed-in user and the workspace this connection can read |
get_workspace | Workspace name, website and reporting settings |
list_campaigns | Tests in the workspace, with status and type |
get_campaign | One test's name, type, status, page URL and dates |
get_campaign_report | Per-metric variant stats, statistical significance and revenue |
get_metric_detail | The full breakdown of one metric in a test |
get_timeseries | Conversion rate, revenue, expected improvement and visitors per day |
get_campaign_timeline | When a test was started, paused, edited or ended |
check_attribution_health | Whether conversions and revenue are being attributed to variants |
get_kpi_overview | Headline numbers across the workspace |
search | Searches tests by name (the shape ChatGPT expects) |
fetch | Returns one search result in full (the shape ChatGPT expects) |
Connect your AI tool
You'll find the same instructions, with copy buttons and one-click install links, in Splitlab under Workspace settings → API keys → Connect AI tools.
Every client below signs in with OAuth: the first time it connects, your browser opens Splitlab, you sign in, choose the workspace and approve.
Claude (claude.ai and Claude Desktop)
- Open Settings → Connectors.
- Choose Add custom connector and paste
https://mcp.splitlab.io/mcp. - Select Connect, then sign in to Splitlab and approve access.
Claude Code
claude mcp add --transport http splitlab https://mcp.splitlab.io/mcpThen start Claude Code and run /mcp to sign in.
ChatGPT
- In ChatGPT settings, turn on Developer mode. It is under Apps & Connectors → Advanced settings, or under Security on some accounts. It needs a paid plan, and on work accounts an admin may have to allow it.
- Choose Create, name it
Splitlab, pastehttps://mcp.splitlab.io/mcpand choose OAuth authentication. - Save, then sign in to Splitlab and approve access.
Cursor
Add Splitlab to Cursor
with one click, or add this to ~/.cursor/mcp.json:
{
"mcpServers": {
"splitlab": {
"url": "https://mcp.splitlab.io/mcp"
}
}
}VS Code (GitHub Copilot)
Add Splitlab to VS Code
with one click, or add this to .vscode/mcp.json in your project:
{
"servers": {
"splitlab": {
"type": "http",
"url": "https://mcp.splitlab.io/mcp"
}
}
}Copilot uses MCP tools in Agent mode.
Gemini CLI
gemini mcp add --transport http splitlab https://mcp.splitlab.io/mcpOr add the server to ~/.gemini/settings.json:
{
"mcpServers": {
"splitlab": {
"httpUrl": "https://mcp.splitlab.io/mcp"
}
}
}Then start Gemini CLI and run /mcp auth splitlab to sign in.
Windsurf
Add this to ~/.codeium/windsurf/mcp_config.json, then refresh the MCP servers in Cascade:
{
"mcpServers": {
"splitlab": {
"serverUrl": "https://mcp.splitlab.io/mcp"
}
}
}Other clients
Point any MCP client that supports Streamable HTTP at https://mcp.splitlab.io/mcp. If it supports
OAuth, it discovers the sign-in flow on its own. If it doesn't, use an API key.
Use an API key instead of OAuth
For clients that can't sign in with OAuth, or for scripts:
- In Splitlab, open Workspace settings → API keys and click Create key.
- Copy the key. It starts with
sl_and is shown once. - Configure your client to send it on every request:
Authorization: Bearer sl_your_key_hereA key reads only the workspace it was created in. Revoke it from the same page and any client using it loses access immediately.
Troubleshooting
| Symptom | What to do |
|---|---|
401 Unauthorized | The sign-in expired or was revoked, or the API key is wrong or revoked. Reconnect the client (in Claude Code, run /mcp; in Gemini CLI, /mcp auth splitlab), or create a new key |
The Splitlab MCP server is not enabled (503) | The MCP server isn't switched on yet. Contact Splitlab support |
429 Too Many Requests | Each credential can make 120 requests per minute, and each network address 300. Wait a minute and try again |
| The AI tool sees the wrong workspace | Each connection reads one workspace. Disconnect and connect again, choosing the other workspace |