> ## Documentation index
> Fetch the complete documentation index at: [llms.txt](../llms.txt)
> Use this file to discover all available pages before exploring further.

# Sleuth MCP server

Connect any MCP client or agent to Sleuth: run research, look up the knowledge base, build Explore datasets, and search the research library, with 52 tools.

The Sleuth MCP server lets Claude, ChatGPT, Copilot, and any agent that speaks the Model Context Protocol use Sleuth. Every project an agent starts is saved to the user's Sleuth account, so they can open it in Studio, review it cell by cell, and keep working.

## Connection details

|   |   |
| --- | --- |
| Server URL | `https://mcp.sleuthapi.com/mcp` |
| Transport | Streamable HTTP |
| Sign-in for people | OAuth 2.1 with PKCE. Clients register with dynamic client registration or a client ID metadata document. No API key needed. |
| Sign-in for headless agents | A Sleuth API key sent as a bearer token. API keys come with Enterprise. [API](api.md) |
| Plans | Every plan, including the free trial, for people who sign in. Jobs that run without a person need an API key, which comes with Enterprise. |
| Directory listings | Coming soon for Claude and ChatGPT. Until then, add Sleuth as a custom connector with the server URL. |

## Add it to a client

Claude Code:

```
claude mcp add --transport http sleuth https://mcp.sleuthapi.com/mcp
```

Clients that read an `mcpServers` config, like Cursor:

```
{
  "mcpServers": {
    "sleuth": { "url": "https://mcp.sleuthapi.com/mcp" }
  }
}
```

The first call opens Sleuth's sign-in in a browser. Step-by-step guides: [Claude](setup-claude.md), [ChatGPT](setup-chatgpt.md), [Copilot](setup-copilot.md), [other tools](setup-other.md).

## The tools

| Group | Tools | What they do | Credits |
| --- | --- | --- | --- |
| [Projects and runs](tools-projects.md) | 9 | Create a project, ask a research question, follow the run, approve plans, read the answer | Research runs use credits |
| [Datasets and charts](tools-datasets.md) | 7 | Read datasets page by page, download CSV, list versions, get charts as specs or SVG | No |
| [Your documents](tools-documents.md) | 5 | Upload files to a project, search them, get download links, delete them | No |
| [Knowledge base lookups](tools-knowledge-base.md) | 15 | Resolve drugs, companies, indications, and trials to Sleuth IDs and follow their links | No |
| [Explore datasets](tools-explore.md) | 4 | Build a filtered dataset from the knowledge base and download it | No |
| [Research library](tools-research.md) | 4 | Search news, literature, filings, patents, and sell-side research, and read passages | No |
| [Onboarding and health](tools-onboarding.md) | 8 | Set up a new account from the AI tool, and check the server | No |

## What it costs

The same as Studio. A research run uses one credit per analytical cell it fills, and the user approves the plan first unless they turned on auto-approve. Lookups, Explore datasets, research library searches, and questions about results Sleuth already built don't use credits. [How credits work](how-credits-work.md)

## What isn't available over MCP yet

- Workflows run in Studio. Agents can ask for the same analysis as a research question. [Prompts to start with](ai-prompts.md)
- Monitoring, sharing, and plan changes are set up in Studio.

## Next

[How a research run works](mcp-run-lifecycle.md) walks through a run from question to dataset. [Errors and limits](mcp-errors.md) lists every limit and the fix for each error. Writing an agent that reads these docs? See [Docs for AI agents](for-agents.md).

## Related

- [How a research run works](mcp-run-lifecycle.md): Create a project, send the question, wait for the status to change, handle approvals and clarifications, then read the datasets and the answer.
- [Sleuth in Claude, ChatGPT, and Copilot](ai-tools-overview.md): Ask in the AI tool you already use, and Sleuth does the work behind the answer.
- [Docs for AI agents](for-agents.md): Every page in these docs has a Markdown version, there’s an index of all of them, and the whole set comes as one file. Here’s what’s where, and how agents should use Sleuth.

*Updated November 9, 2026.*
