---
title: "MCP for authors"
url: "/help/buuk/mcp-for-authors"
updated: "2026-10-22T11:13:51.070Z"
---
# MCP for authors

The authoring MCP at `/api/mcp` puts your docs where your developers already work. It answers the same JSON-RPC methods as the reader endpoint — `initialize`, `tools/list`, `tools/call` — but you authenticate with a `buuk_cli_` token (or a signed-in dashboard session) instead of connecting publicly.

Send the token as a bearer header on every call: `Authorization: Bearer buuk_cli_…`. Each token is scoped to one workspace, and a read-only token is enough for everything — the whole server is read-only.

## Connect Cursor or Claude Code

<!-- Cursor -->

```json
// .cursor/mcp.json in your project
{
  "mcpServers": {
    "buuk": {
      "url": "https://app.getbuuk.com/api/mcp",
      "headers": {
        "Authorization": "Bearer buuk_cli_..."
      }
    }
  }
}
```

<!-- Claude Code -->

```bash
claude mcp add --transport http buuk https://app.getbuuk.com/api/mcp \
  --header "Authorization: Bearer buuk_cli_..."
```

> [!TIP]
> Keep the token in a secret store, such as `BUUK_API_TOKEN` in your environment — never in the repo. See [API tokens and MCP](./api-tokens-and-mcp) for creating and revoking tokens.

## What it exposes

It lists the same seven tools as the public server: `search_docs`, `get_article`, `list_sections`, `list_recent_updates`, `list_api_specs`, `search_api_reference`, and `get_api_operation`. An agent can find topics, read published text as markdown, and pull exact wording from your API reference without leaving its environment.

Read-only on purpose: an agent connected this way cannot edit or publish. If a job needs machine writes, the public API is the path — creating articles from markdown is a **Can write** REST call, not an MCP tool.

## Team workflows

- **Search before you write. **Run `search_docs` to see whether a topic already has a home before drafting a new article.
- **Draft updates from the live text. **Fetch a page with `get_article`, have the agent revise it, and bring the result back into the dashboard editor. A person reviews and publishes.
- **Check what changed. **Call `list_recent_updates` to see which pages shipped lately and how fresh the docs are.
- **Keep API answers exact. **`search_api_reference` and `get_api_operation` hand the agent real endpoint details from your published reference.

> [!NOTE]
> Nothing auto-publishes. The agent edits a copy, a teammate reviews the result, and publishing stays a human step.

## Related

- Mint and manage the token itself: [API tokens and MCP](./api-tokens-and-mcp).
- The public, token-free version readers’ tools use: [MCP for readers](./mcp-for-readers).
- The CLI commands behind the same token: [CLI reference](./cli-reference).
- Where a revised draft becomes a live page: [Write and publish an article](./write-and-publish-an-article).