# Connect an agent

Source: https://docs.bearletter.com/mcp/connect

Bearletter is a remote MCP server. Agents that speak MCP (Claude, Codex, Cursor and others) can read and change your blog with the same rules as the dashboard.

```text
https://api.bearletter.com/mcp
```

## Let your agent do it

Agents that can run commands or edit their settings add Bearletter themselves; chat apps tell you where to click.

```text
Set up Bearletter for me and help me work on my blog.

1. Read https://docs.bearletter.com/llms.txt, then open only the pages you need. Every page is Markdown.
2. Know the blog's rules before writing: the site files have fixed names (home.md, nav.md, directories.md, redirects.md, social.md, blogroll.md, footer.md, post-header.md, post-footer.md, post-aside.md, styles.css, podcast.md) that you never rename or add to; only files under posts/ and pages/ are yours to name.
3. Add the Bearletter MCP server to your own configuration: a remote (HTTP) server named bearletter at https://api.bearletter.com/mcp. Use your own way of adding MCP servers (a command or your config file). If you cannot change your own settings, tell me exactly where I add it. There is no key to copy.
4. If you take skills, also install the Bearletter skill: https://docs.bearletter.com/skills/bearletter/SKILL.md (how: https://docs.bearletter.com/mcp/skills.md).
5. When you first use the server, it opens a Bearletter page in my browser. I will sign in and approve it.
6. Check the connection by calling get_blog, and tell me my blog's title and address.
7. From then on, read before you change anything, tell me what you will change, and wait for my yes before saving anything that goes public. If a save is refused, fix the lines it names and try again.
```

## Or add it yourself

Any agent that supports the Model Context Protocol works: add the address above as a **remote MCP server** (HTTP). There is no key to copy. The first time the agent uses Bearletter, it finds the sign-in page by itself and opens it in your browser.

| Agent | How to add it |
| --- | --- |
| Claude Code | `claude mcp add --transport http bearletter https://api.bearletter.com/mcp`, then `/mcp` to sign in |
| Claude (web and desktop) | Customize › Connectors › Add custom connector |
| ChatGPT | Settings for apps and MCP servers › add a custom MCP server, with OAuth |
| Codex | `[mcp_servers.bearletter]` with `url = "https://api.bearletter.com/mcp"` in `~/.codex/config.toml`, then `codex mcp login bearletter` |
| OpenCode | `"mcp": { "bearletter": { "type": "remote", "url": "https://api.bearletter.com/mcp" } }` in `opencode.json` |
| Cursor | `"mcpServers": { "bearletter": { "url": "https://api.bearletter.com/mcp" } }` in `~/.cursor/mcp.json` |
| VS Code | `"servers": { "bearletter": { "type": "http", "url": "https://api.bearletter.com/mcp" } }` in `.vscode/mcp.json` |
| Gemini CLI | `gemini mcp add --transport http bearletter https://api.bearletter.com/mcp` |

For the technically curious: Bearletter answers an unauthenticated request with the standard pointer to its OAuth authorization server, and accepts both client metadata documents and dynamic client registration, so agents need nothing configured beyond the address.

## Approve it

The first time the agent uses Bearletter, it opens a Bearletter page in your browser. Sign in, check what the agent asks to do, untick anything you do not want, and choose **Allow**. Reading your subscribers' email addresses is off unless you tick it.

| Permission | Lets the agent |
| --- | --- |
| `read` | Read your blog, drafts and stats |
| `write` | Write, publish and delete posts, pages and episodes |
| `media` | Upload files to your media library |
| `blog` | Change your blog's home page, menu, theme and settings |
| `subscribers` | See and export your subscribers' email addresses |

Changing the blog from an agent needs Pro, as in the dashboard. If the agent later needs a permission you left out, it asks, and you approve it the same way.

## Check it

Ask the agent to call `get_blog`. It answers with your blog's title and address.

## Stay in control

- The agent's access lasts until you remove it: **Settings › Apps & tokens** shows every connected assistant, with **Disconnect**.
- Access renews itself while you use it, and ends after 30 days without use.
- Anything an agent publishes is live at once, like in the dashboard. Good agents tell you what they will change first; the prompt on the [home page](/) asks them to.
