MCP Server

Flint Pay runs a Model Context Protocol (MCP) server for the developer docs. Point an MCP-capable AI agent at it and the agent can search the docs, read any page, and pull full request/response schemas for any endpoint on demand, so it works from the current API surface instead of guessing from memory.

The server is live at:

https://developers.withflintpay.com/mcp

It is read-only and covers documentation only. It does not touch your Flint account, your data, or the payments API, so no Flint API key is involved.

When to use this#

The MCP server and SKILL.md solve the same problem (giving an agent correct Flint context) in two different ways:

SKILL.md is one dense file you paste into the agent's context up front. Best when you want the whole surface loaded at once, offline, with no network calls mid-task.
The MCP server is a live connection the agent queries as it works. Best when you want the agent to pull only the pages and schemas it needs, always from the current docs, without spending context on the parts it does not.

Use SKILL.md for a fixed reference, the MCP server for interactive lookups. They are not mutually exclusive.

Tools#

The server exposes four tools:

Tool	What it does
`search_docs`	Search guides, API reference, and blog. Returns matching pages with URLs.
`get_page`	Fetch one docs page as Markdown by its path (for example `/docs/guides/webhooks`).
`list_endpoints`	List public API endpoints (method, path, summary), optionally filtered by resource (for example `orders`).
`get_endpoint`	Get the full reference for one endpoint by its `operationId`, including request and response schemas and code samples.

A typical agent loop is search_docs or list_endpoints to find the right operation, then get_endpoint to read its exact schema before writing the call.

Connect an agent#

Add the server with the Claude Code CLI:

Bash

claude mcp add --transport http flint-docs https://developers.withflintpay.com/mcp

Then ask Claude to work against the Flint API. It will call the Flint docs tools when it needs endpoint details.

The transport is stateless JSON-RPC 2.0 over HTTP POST. List the tools directly:

Bash

curl -s https://developers.withflintpay.com/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Call one:

Bash

curl -s https://developers.withflintpay.com/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"search_docs","arguments":{"query":"idempotency"}}}'

Protocol notes#

Transport. Stateless Streamable HTTP: JSON-RPC 2.0 over POST. There is no session or server-initiated stream, so a GET returns 405.
Protocol version. The server advertises 2025-03-26 and also accepts 2024-11-05 and 2025-06-18 clients.
Batching. Batched requests are supported, capped at 20 messages per call.
Auth. None. The endpoint is public and read-only.

The MCP server reads the same docs corpus as this site, so it always reflects the current API surface. The full corpus is also available as a single file at /llms-full.txt.