# Bolt MCP server

> Connect an agent to Bolt Help over MCP to search and read documentation on demand.

Source: https://help.boltapp.com/ai-agents/mcp-server

Model Context Protocol (MCP) lets an agent search and read documentation as it works, instead of loading the whole site into its context. The Bolt documentation server answers keyword searches and returns individual pages as Markdown.

## Endpoint [#endpoint]

```text title="Endpoint"
https://help.boltapp.com/mcp
```

No authentication. This server reads published documentation and nothing else, so there is no key to configure and no account to connect.

This is not the merchant MCP server, which reads a merchant's own Bolt account over OAuth. The two are separate, and this one cannot see account data.

## Tools [#tools]

| Tool                | Input                                          | Returns                                                    |
| ------------------- | ---------------------------------------------- | ---------------------------------------------------------- |
| `search_docs`       | `query`, optional `limit` (1 to 25, default 8) | Ranked pages with title, path, description, and an excerpt |
| `get_doc_page`      | `path`, such as `/getting-started/api-keys`    | The full page as Markdown                                  |
| `list_doc_sections` | None                                           | The documentation sections and their browse anchors        |

`search_docs` covers guides and the API reference. An endpoint page returned by `get_doc_page` carries the OpenAPI definition of that endpoint, with every schema it references included.

## Set up your client [#set-up-your-client]

<CodeGroup>
  <CodeBlockTabs defaultValue="Terminal">
    <CodeBlockTabsList>
      <CodeBlockTabsTrigger value="Terminal">
        Terminal
      </CodeBlockTabsTrigger>

      <CodeBlockTabsTrigger value=".cursor/mcp.json">
        .cursor/mcp.json
      </CodeBlockTabsTrigger>

      <CodeBlockTabsTrigger value=".vscode/mcp.json">
        .vscode/mcp.json
      </CodeBlockTabsTrigger>
    </CodeBlockTabsList>

    <CodeBlockTab value="Terminal">
      ```bash
      claude mcp add --transport http bolt-docs https://help.boltapp.com/mcp
      ```
    </CodeBlockTab>

    <CodeBlockTab value=".cursor/mcp.json">
      ```json
      {
        "mcpServers": {
          "bolt-docs": {
            "url": "https://help.boltapp.com/mcp"
          }
        }
      }
      ```
    </CodeBlockTab>

    <CodeBlockTab value=".vscode/mcp.json">
      ```json
      {
        "servers": {
          "bolt-docs": {
            "type": "http",
            "url": "https://help.boltapp.com/mcp"
          }
        }
      }
      ```
    </CodeBlockTab>
  </CodeBlockTabs>
</CodeGroup>

In Claude Code, run the command above and confirm with `claude mcp list`. In Cursor and VS Code, save the file in your project and reload the window, then check that `bolt-docs` lists three tools.

For Claude desktop and web, add it under **Settings**, **Connectors**, **Add custom connector**, and paste the endpoint.

In ChatGPT, remote MCP servers are apps, and adding one requires developer mode. Give ChatGPT the trailing slash, `https://help.boltapp.com/mcp/`: ChatGPT reads the server's tools once when you create the app, and that one read is the request least able to tell you it followed a redirect. Who can add an app depends on the plan:

| Plan               | Who can add the app                                                                                                                               |
| ------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------- |
| Business           | Admins and owners only, under **Workspace settings**, **Apps**, **Create**                                                                        |
| Enterprise and Edu | Whoever an admin has granted developer mode, under **Settings**, **Apps**, **Advanced settings**, or **Workspace settings**, **Apps**, **Create** |
| Pro                | Any Pro user, in developer mode, for read and fetch actions only. Full MCP support is Business and Enterprise or Edu only                         |

After you add the app, select it from the tools menu or @-mention it in a prompt when you want an answer from current Bolt documentation. Deep research can use the app for read and fetch actions: the server's tools do not need to be named `search` and `fetch`. See [developer mode and MCP apps in ChatGPT](https://help.openai.com/en/articles/12584461-developer-mode-and-mcp-apps-in-chatgpt).

## Verify the connection [#verify-the-connection]

Ask your agent to search rather than guess:

```text title="Example prompt"
Use the bolt-docs MCP server to find how to verify a Bolt webhook signature, then open the page it returns.
```

A working connection returns the webhook verification page. If the agent answers from memory instead, the server is not connected: check that your client lists three tools under `bolt-docs`.

## Keep the path [#keep-the-path]

The endpoint path is `/mcp`, and a bare `https://help.boltapp.com` does not serve MCP. A request to `/mcp` is redirected to `/mcp/`, which every client tested here follows without losing the request. If your client reports a redirect it will not follow, configure `https://help.boltapp.com/mcp/` instead.

## Other ways to load documentation [#other-ways-to-load-documentation]

An MCP server suits an agent that searches as it works. For an agent you want to prime up front, [llms-full.txt](https://help.boltapp.com/llms-full.txt) is every prose page in one file, and [Docs for agents](/ai-agents/docs-for-agents) covers that alongside the `llms.txt` index, per-page Markdown, and the published OpenAPI specs.

## Related [#related]

* [Docs for agents](/ai-agents/docs-for-agents)
* [Skills](/ai-agents/skills)
* [Integration prompts](/ai-agents/integration-prompts)