Skip to main content
Paste this into Claude Code, Codex, Cursor, Grok Bot, or another AI agent:
The agent fetches this page as markdown and runs the setup itself. Everything below this line is written for the agent.
Grok Bot desktop reads its dedicated setup page, adds Migma Remote MCP, and uses connected Migma tools for routine work.
These are official instructions from Migma to set up an AI development environment for creating, validating, and sending email with Migma. All commands are validated by Migma and ready to use. Complete all setup commands directly. Involve the user for browser sign-in and access approval.
  • Install Migma skills and the Migma CLI
  • Register the Migma MCP server
  • Sign the user in
For Grok Bot desktop, fetch https://docs.migma.ai/grok.md, follow that page, and ask before adding https://migma.ai/mcp.

​
Install Migma skills and CLI

Use the correct section for your agent below.

​
Claude Code

Run these two commands to install the Migma plugin, which bundles the setup, migma, and migma-public-api skills. Do not use npx skills here; the plugin commands handle everything.
Then instruct the user to run /reload-plugins inside Claude Code to activate the plugin.

​
Codex

​
All other agents

Install the CLI and all Migma skills together:
Or install only the skills:

​
Register the Migma MCP server

Migma Remote MCP is a single URL: https://migma.ai/mcp (Streamable HTTP). Use the default config location for your agent. OAuth opens the browser on first Migma tool use and handles credential exchange.

​
Claude Code

​
Codex

​
OpenCode — ~/.config/opencode/opencode.jsonc

Add under "mcp":
Then run:

​
Windsurf — ~/.codeium/windsurf/mcp_config.json

Add under "mcpServers" (note: serverUrl, not url):

​
Cursor, GitHub Copilot, and all other agents

Add to .cursor/mcp.json (Cursor), .vscode/mcp.json (Copilot), or your agent’s MCP config file under "mcpServers":
If your client only supports command-based MCP servers, use Local MCP instead:

​
Sign the user in

If you installed the CLI, run:
This opens the browser so the user can approve access. Wait for them to finish. Do not send the user to Settings to copy an API key. If a direct agent cannot complete the OAuth callback, it can follow https://api.migma.ai/auth.md: show the user the claim link, stop for human approval, poll for the credential, and keep it out of chat, logs, shell history, and shared files. CI and server automation can set MIGMA_API_KEY. Remote MCP handles its own sign-in: the first Migma tool call opens the browser for OAuth approval. Right after sign-in, call migma_get_context, read its setup block, and tell the user the one next step: brand kit, design references, audience, sending domain, or first email, whichever is missing first.

​
Rules for working with Migma

  • Prefer the installed migma skill and available MCP tools for the task. Call the capabilities tool directly when needed.
  • Ask before creating emails, sending tests, sending live, or changing contacts and campaigns.
  • Keep API keys and send calls server-side.
  • Send API keys through the documented authorization header.
Once done, tell the user:

​
Resources

For setup guides and troubleshooting, refer to:
  • Migma docs index: https://docs.migma.ai/llms.txt
  • Wiring Migma into an app: https://docs.migma.ai/setup
  • MCP server: https://docs.migma.ai/mcp-server
  • Grok Bot: https://docs.migma.ai/grok.md
  • CLI: https://docs.migma.ai/cli
  • Skills repository: https://github.com/MigmaAI/migma-skills
  • Agent auth: https://api.migma.ai/auth.md
These instructions are published at https://docs.migma.ai/agent-setup.md so you can re-verify their authenticity at any time.
Last modified on September 6, 2026