Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.gooseworks.ai/llms.txt

Use this file to discover all available pages before exploring further.

MCP (Model Context Protocol) servers let you connect external tools and services to your agent, giving Goose access to new capabilities.

​
What is MCP?

MCP is an open protocol that allows AI agents to use external tools. When you connect an MCP server to Goose, it gains access to the tools that server provides — like reading from a database, managing a project board, or querying an API.

​
GooseWorks MCP server

GooseWorks ships an MCP server that you can plug into any MCP-aware coding agent (Claude Code, Cursor, etc.). It gives your coding agent direct tool access to three parts of your GooseWorks account: your files, your chat history, and your automations.
MCP vs CLI. The GooseWorks MCP and the GooseWorks CLI are two different things with different uses:
  • The MCP gives your coding agent tools for your files, chat history, and automations. It does not include the skill catalog.
  • The CLI installs a master skill that gives your coding agent access to the 100+ GooseWorks skill catalog (scraping, enrichment, research, etc.). The CLI does not give access to files, chats, or automations.
You can install one or both. npx gooseworks install --mcp (or --all) is a convenience that installs the CLI skill and registers this MCP server in one command. See CLI.

​
GooseWorks MCP — gooseworks

  • URL: https://mcp.gooseworks.ai/mcp
  • Transport: HTTP (streamable)
  • Auth: Authorization: Bearer <your-api-key>
Exposes three tool families, all scoped to the agent workspaces and shared organization folders the calling user has access to.

​
Files

Operate on files in your agent workspaces or shared org folders.
  • list_accessible_scopes — list every workspace you can reach
  • list_directory — browse a workspace one level deep
  • read_file — fetch a file inline (up to 10MB; binary returned as base64)
  • write_file — create or overwrite a file via JSON (up to 50MB)
  • get_upload_url / get_download_url — presigned S3 URLs for large or binary payloads
  • delete_file — remove a file or directory recursively
  • search_files — find files by path substring

​
Chat history

Search and read what your agents have already discussed.
  • search_chat_history — search across chat sessions by keyword
  • list_chat_sessions — list recent chat sessions for an agent
  • get_chat_transcript — fetch the full message transcript for a session

​
Automations

Schedule and trigger your agent from the outside world. Your coding agent can create and manage these on your behalf.
  • list_automations / get_automation — inspect existing automations
  • create_automation — schedule a recurring job, or wire up an inbound webhook or email trigger
  • update_automation — change the cron schedule, trigger config, or the agent instructions
  • set_automation_enabled — pause or resume without deleting
  • delete_automation — remove it
Chat and automation tools target a single agent at a time. File tools can additionally target shared org folders.

​
Installing via the CLI

The GooseWorks CLI registers the GooseWorks MCP server in your coding agent’s config for you. By default, npx gooseworks install only installs the GooseWorks skill — the MCP is opt-in via a flag: 
# GooseWorks MCP
npx gooseworks install --cursor --mcp

# Everything, all detected agents
npx gooseworks install --all
See CLI → Install flags for the full matrix.

​
Manual setup (without the CLI)

Prefer to wire up MCP by hand? You need an API token, then you drop it into your MCP client’s config.

​
Step 1 — Create an API token

  1. Go to app.gooseworks.ai/admin/settings/api-tokens.
  2. Click Create MCP Token. This creates a user-scoped token with the mcp:access scope.
  3. Copy the token (starts with cal_...) immediately — it’s shown once.
Why user-scoped? A user-scoped token is not pinned to a single agent. The MCP lets the agent pick a workspace target per tool call. If you use an agent-scoped token, you’ll be stuck on one workspace.

​
Step 2 — Add to your MCP client

{
  "mcpServers": {
    "gooseworks": {
      "type": "http",
      "url": "https://mcp.gooseworks.ai/mcp",
      "headers": {
        "Authorization": "Bearer cal_YOUR_TOKEN"
      }
    }
  }
}
The gooseworks-app/.cursor/mcp.json (project-level) file works the same way if you’d rather scope the config to one project.

​
Step 3 — Restart your editor

Fully quit and reopen Cursor / Claude Code so it re-reads the config. You should see the server listed as connected in the MCP panel.

​
Rotating or revoking

  • Rotate: create a new token, update mcp.json, revoke the old one in the UI.
  • Revoke: Settings → API Tokens → trash icon next to the token. Revocation is immediate — all active MCP sessions using that token will fail on their next request.

​
Adding your own custom MCP servers

You can also connect third-party MCP servers to Goose:
  1. Go to Settings > MCP Servers.
  2. Click Add Server.
  3. Enter the server URL and transport type (HTTP, SSE, or stdio).
  4. Goose will discover the available tools automatically.

​
One-click integrations

Gooseworks also supports one-click integrations with popular services through OAuth:
  • Notion
  • GitHub
  • Linear
  • Google Docs
  • Google Sheets
  • Jira
  • Confluence
  • Gmail
  • Slack
These integrations connect with a single click and give Goose access to read and write data in those services.