> For the complete documentation index, see [llms.txt](https://ascribed-dev.com/docs/llms.txt). This documentation describes Ascribe 0.4.4. Anything marked in main, not yet released isn't in that release yet.

# Set up the MCP server

> How-to · Give an agent with no shell Ascribe's answers as tools, through ascribe mcp.

`ascribe mcp` is a server for the [Model Context Protocol](https://modelcontextprotocol.io), which most agents' hosts speak. It offers the commands that answer questions as tools an agent calls, for an agent with no shell, or a host that asks you to approve each command it runs. An agent with a shell needs no server: the commands give the same answers.

In VS Code, the extension offers the server itself, so there's nothing to set up: see [Agents in the editor](https://ascribed-dev.com/docs/editor/agents.md#agents-in-vs-code). In Claude Code, [the plugin](https://ascribed-dev.com/docs/agents/claude-code.md) has it, and `ascribe agents sync --target claude --with-hook` writes it into `.mcp.json`.

## Add the server to your agent's host

Your agent's host starts the server, in the repository's root, and one server serves every project in it.

1. Open the file your host reads its servers from, such as `.mcp.json` at the repository's root for Claude Code, or `.cursor/mcp.json` for Cursor.

2. Add Ascribe to it:

   ```json
   {
     "mcpServers": {
       "ascribe": {
         "command": "npx",
         "args": ["ascribe", "mcp"]
       }
     }
   }
   ```

3. Start a new session, and check that the host lists the server's tools.

## What it offers

Its tools are the commands that answer questions; its resources are each directive's syntax, and each project's [content model](https://ascribed-dev.com/docs/glossary.md#content-model) and rules; and its prompts are `new-page`, `fix`, and `review`, which many hosts show as slash commands. Every tool is read-only, so `ascribe_check` never runs [Vale](https://ascribed-dev.com/docs/checks/vale.md): `ascribe check --vale` in a shell, which your host asks you about, checks the prose. A tool's result is its command's JSON, whose schema is in the [JSON report contract](https://ascribed-dev.com/docs/contracts/json-reports.md), and no tool gives an output schema, which would make the tool list five times as long. [`ascribe mcp`](https://ascribed-dev.com/docs/agents/commands.md#ascribe-mcp) lists each tool and the command it runs, and the resources.