Skip to Content
MCP serverSetup

Setup

Connecting the Traveler.md MCP server is a one-time, two-step process: add the server URL to your MCP client, then complete the OAuth flow when prompted.

What you need

  • A TravelAI account. Sign up at traveler.md  if you don’t have one.
  • An MCP-aware client. The most common are listed below; the underlying flow is the same in any client that supports remote MCP servers with OAuth.

Install

The fastest path is to follow the guide for the app you’re using. Each one has the exact steps, plan requirements, and the assistant instruction block to paste in once you’re connected:

Agent Plugin

If your client supports Agent Plugins , you can install the server and its usage guide together instead of adding the URL by hand:

github.com/TravelAiSolution/traveler-md-agent-plugin 

The package bundles the server configuration with an Agent Skill that teaches your client how to use the tools well: read before you write, how a section is replaced rather than appended to, the per-section limits, and how to recover from each error the server returns. Follow your client’s instructions for adding a Git-hosted Agent Plugin and point it at that repository.

Agent Plugins is a new standard, so support varies between clients. If yours doesn’t mention plugins yet, use one of the guides above or the server URL below.

Any other MCP client

Any client that speaks the MCP streamable-HTTP transport works. If yours isn’t in the list above, most only need the server URL — they discover the OAuth endpoints automatically:

  • Server URL: https://mcp.traveler.md/mcp
  • Auth type: OAuth 2.1 with PKCE

If your client asks for the OAuth endpoints by hand, they live on connect.traveler.md:

  • Authorization endpoint: https://connect.traveler.md/v1/oauth/authorize
  • Token endpoint: https://connect.traveler.md/v1/oauth/token

See Authentication for the full endpoint list.

See Set up Traveler.md in another MCP client for the generic walkthrough.

Verify

Once connected, ask your agent something like “What do you know about my travel preferences?” If the connector is wired up correctly, the agent will call the read_profile tool and respond with what’s in your file.

If you get an error, jump to Debugging.

Make it stick

Connecting the server makes the tools available. It doesn’t guarantee your assistant reaches for them, and the turn where that matters most is the one where it feels least necessary: asked “where should we stay in Lisbon?”, an assistant will happily answer from the conversation alone and never open your profile.

Each client guide ends with an instruction block for exactly this. Where you paste it decides whether it lasts:

  • Put it somewhere persistent if your client has one — Claude Project instructions, ChatGPT custom instructions, a CLAUDE.md or AGENTS.md, a Cursor project rule. Then it applies to every future conversation.
  • Pasting it into a chat works too, but only for that chat. The next conversation starts without it.

Clients that support Agent Plugins  get this from the bundled skill and need no manual step.

A second check is worth doing after that: ask for a recommendation rather than for your preferences, something like “suggest a hotel for my next trip.” A correctly configured assistant reads your profile and the trip first, then makes its own suggestion. If it recommends without reading anything, the instruction block isn’t in a place the assistant sees.

What happens during connection

  • Your client requests authorization for a set of scopes.
  • TravelAI shows you a consent screen listing what the client will be allowed to read or write.
  • After you allow, your client receives an access token and refresh token. The tokens are stored by the client, not by TravelAI.
  • You can revoke access at any time from Connections at traveler.md .