> ## Documentation Index
> Fetch the complete documentation index at: https://confidence.spotify.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Experiments MCP Server

> Reference for the Confidence Experiments MCP server: endpoint, authentication, and the 6 read-only tools for finding experiments and reading their results.

The Experiments MCP server lets an AI assistant find A/B tests and rollouts and read their configuration and results. All its tools only read data. You can't create or launch an experiment through MCP today. To do that with an assistant, use the in-app Confidence Agent. See [Work with the Confidence Agent](/docs/guides/work-with-the-confidence-agent).

To connect the server, see [MCP Servers](/docs/reference/agent/mcp-servers#integration-setup).

<Warning>
  The Confidence MCP servers are experimental. Tools and their parameters can change.
</Warning>

## Endpoint and Authentication

Endpoint: `https://mcp.confidence.dev/mcp/experiments`. Transport: streamable HTTP. Authentication: sign in to Confidence from your AI assistant. The server registers 6 read-only tools for A/B tests and rollouts.

## Experiments

| Tool | Description | Key parameters |
| - | - | - |
| `list_experiments` | Lists A/B tests and rollouts with names, display names, states, owners, and creation times. In list mode, use `filter` and `orderBy`. In search mode, pass `query` for full-text search ordered by relevance. Search mode ignores `filter` and `orderBy`. | `filter`, `orderBy`, `query`, `pageToken` |
| `get_experiment` | Returns a compact summary of an experiment: name, state, owner, metrics, decision, and analysis result names. Set `summary=false` for configuration, checks, stats, and state history. | `experimentName`, `summary` |
| `get_results` | Returns statistical results: relative effect estimates, confidence intervals, and status messages per metric. Accepts an instance name, such as `workflows/abtest/instances/{id}`, or an analysis result name. Set `summary=false` for full output. | `analysisResultName`, `summary` |

The resource name shows the experiment type, for example `workflows/abtest/...` or `workflows/rollout/...`.

## Resources

| Tool | Description | Key parameters |
| - | - | - |
| `get_resource` | Returns a resource referenced in experiment data. Supports `metrics/*`, `entities/*`, `surfaces/*`, `segments/*`, `factTables/*`, `flags/*`, `identities/*`, `assignmentTables/*`, and `clients/*`. | `resourceName` |
| `list_entities` | Lists entities in the account. | `pageToken` |
| `list_fact_tables` | Lists fact tables with their entity columns, timestamp columns, measures, and dimensions. | `pageToken` |