Skip to main content

MCP servers

Flynet ships two MCP servers. They do different jobs and run side by side:
  • Docs MCP — your agent searches the live Flynet docs while it works. Hosted by Mintlify, zero setup, no credentials.
  • API MCP — an npm package your agent runs locally to call the Flynet API through read-only tools, with your own credentials. Nothing to host.

Docs MCP

The docs site hosts its own MCP server, so any MCP-aware agent can search Flynet’s documentation in real time instead of working from a stale copy. There’s nothing to deploy and no key to manage.
No MCP client? Every page is available as markdown for any model: llms.txt indexes the docs, llms-full.txt is the whole site in one file, and appending .md to any docs URL returns its source.

API MCP

The API MCP is a published npm package, @flynetdev/mcp — there’s no server to connect to and nothing to host. Your MCP client runs it locally with npx, and you supply your own credentials in the client’s config, so no secrets leave your machine. It gives your agent 9 read-only tools that call Flynet directly — fetch a restaurant, list venues, or read the signed-in member’s wallets while it reasons, instead of writing throwaway fetch() code. It wraps @flynetdev/core, the same client you’d use in production.

Credentials

Flynet has two auth schemes, so the package takes two credentials and routes each tool to the right one. Provide whichever surfaces you’ll use: Both are server-side secrets — keep them on your machine, never in client-side code. Get them from sandbox access.

Install

Tools

Every tool is read-only. List tools are paginated and return trimmed fields; pass include_raw: true for the full object. When a tool is called without the credential it needs, it returns a clear message naming the missing variable and the fix — not a stack trace.

Next

Install the agent skill →

How it fits together →