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

# The Tatara MCP server

> What the Tatara MCP server exposes, the fifteen tools an agent can call, and how rejections guide an agent to correct itself.

Tatara exposes your company brain to agents over the Model Context Protocol. The server speaks JSON-RPC over HTTP at a single endpoint:

```text theme={null}
https://www.tatara.app/api/mcp
```

Any MCP client with HTTP transport can connect. See [Connecting an agent](/mcp/connect) for configuration and sign-in.

## What an agent can do

Fifteen tools cover the life of a document: find it, read it, write it, file it, and get it back after a delete.

### Reading

| Tool               | What it does                                                                         |
| ------------------ | ------------------------------------------------------------------------------------ |
| `get_taxonomy`     | Returns the brain's folders, document types, tag registry, and available formats.    |
| `get_document`     | Reads one document by `id` or `path`, with its outbound links and inbound backlinks. |
| `search_documents` | Full-text search across titles, descriptions, and bodies, with ranked snippets.      |
| `list_documents`   | Browses and paginates documents without returning their content.                     |

### Writing

| Tool                   | What it does                                                |
| ---------------------- | ----------------------------------------------------------- |
| `write_document`       | Creates a document, or updates any part of an existing one. |
| `update_document`      | Applies ordered string replacements to a document body.     |
| `create_html_document` | Creates a self-contained static HTML document.              |

### Organizing

| Tool            | What it does                                             |
| --------------- | -------------------------------------------------------- |
| `create_folder` | Creates one folder, optionally under an existing parent. |
| `move_document` | Moves a document to another folder or to the top level.  |
| `move_folder`   | Moves a folder and everything inside it.                 |
| `rename_folder` | Renames a folder and rewrites the paths beneath it.      |

### Deleting and restoring

| Tool               | What it does                                                        |
| ------------------ | ------------------------------------------------------------------- |
| `delete_document`  | Soft-deletes a document.                                            |
| `delete_folder`    | Soft-deletes a folder, and its contents when you confirm a cascade. |
| `restore_document` | Brings back a deleted document.                                     |
| `restore_folder`   | Brings back a deleted folder and everything deleted inside it.      |

The tools describe themselves. Each one carries its own documentation, and `get_taxonomy` returns your brain's folders, types, tag registry, and a canonical example of every document format — so a connected agent can discover everything it needs at the moment it needs it. The [agent skills](/skills) teach the full write workflow up front, which saves the discovery round trips.

## How rejections work

Validation on every write path is strict, and every rejection states two things: what went wrong, and how to fix it — because the agent, not a person, is the one who has to act on it. An unknown format lists the valid ones. An ambiguous folder name lists the candidate paths and names the field to resend. A body that parses but is not in the format's exact spelling comes back with the corrected bytes attached, so the retry is mechanical rather than a guess.

That way agents correct themselves inside their own loop, instead of leaving a half-written document or a board with no columns for a person to find later. A rejected write is a normal step in a working agent loop, not a fault.

Two behaviors are worth knowing as a person:

* An agent asked to delete a non-empty folder is told how much would go, and told to confirm with you before anything is removed.
* When a workspace reaches its plan's document cap, agent writes are blocked with the same explanation you would see. See [Plans and limits](/account/plans).

<Columns cols={3}>
  <Card title="Connect an agent" icon="plug" href="/mcp/connect">
    Sign in from your client, or use an access token.
  </Card>

  <Card title="Agent skills" icon="code" href="/skills">
    Skills that teach an agent these tools and the document grammars.
  </Card>

  <Card title="Document formats" icon="table" href="/formats/overview">
    What each format is for, and what you can do with it in the app.
  </Card>
</Columns>
