---
title: "Connect your coding agent"
description: "Give Claude, ChatGPT, Cursor and other MCP clients access to your MySpec projects and spec files."
---

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

# Connect your coding agent

MySpec speaks the [Model Context Protocol](https://modelcontextprotocol.io)
(MCP), so any MCP client can list your projects, read and update spec files,
add comments and even start new spec sessions.

There are two MySpec MCP servers. They expose nearly the same tools (see
[the differences](/mcp/remote-server/#how-it-differs-from-the-local-server)); pick
the one that fits where your client runs.

| | Remote server | Local server |
|---|---|---|
| What it is | A hosted endpoint at `https://mcp.myspec.dev/mcp` | The `@myspec/mcp-server` npm package, run with `npx` |
| Transport | Streamable HTTP | stdio |
| Sign-in | OAuth (connectors) or an API token | Browser sign-in or an API token |
| Install | Nothing | Node.js 22 or later |
| Works in | Claude (web, desktop, mobile), ChatGPT, Claude Code, any HTTP MCP client | Claude Desktop, Claude Code, Cursor, VS Code and other local clients |
| Local files | No | Yes: download spec files, upload attachments from disk, share a folder for Brownfield projects |

## Which one should I use?

- **Using Claude or ChatGPT in the browser or the desktop app?** Use the
  [remote server](/mcp/remote-server/) as a connector. You sign in once and
  choose what the app may do.
- **Working in a coding agent on your own machine?** Either works. The
  [local server](/mcp/local-server/) can also save spec files to disk and share
  your codebase with Brownfield sessions.
- **Running on a server, in a container or in CI?** Use an
  [API token](/mcp/api-tokens/) with either server. No browser needed.

## What your agent can do

Each server offers tools in nine groups:

| Group | Examples |
|---|---|
| Projects | `list_projects`, `get_project`, `create_project`, `archive_project` |
| Spec files | `list_spec_file`, `read_spec_file`, `upload_spec_file`, `update_spec_file` |
| Trash | `move_spec_file_to_trash`, `restore_spec_file_from_trash` |
| Attachments | `list_attachments`, `read_attachment`, `upload_attachment` |
| Sessions | `list_spec_sessions`, `start_spec_session`, `archive_spec_session` |
| Artifacts | `list_artifacts`, `create_artifact`, `write_artifact_revision` |
| Stream tokens | `create_stream_token`, `list_stream_tokens`, `revoke_stream_token` |
| Comments | `list_spec_file_comments`, `create_spec_file_comment_thread`, `reply_spec_file_comment_thread` |
| Tasks | `list_tasks`, `update_tasks`, `set_task_status` |

> **People resolve comments, not agents**
>
> There is no tool to resolve, reopen or re-anchor a comment thread. Agents can
> read, add and reply to comments, but deciding that a review is finished stays
> with people.

Source: https://docs.myspec.dev/mcp/index.mdx