FeexPay Docs MCP

Ask your AI assistant to integrate FeexPay, and it answers with the real endpoints instead of making them up.

English · Français

claude mcp add feexpay-docs -- npx -y github:misterkante/feexpay-docs-mcp

See it in action

You: Add MTN Mobile Money payments in Benin to my backend.

Without this server, your assistant fetches docs.feexpay.me, gets an empty JavaScript shell and guesses. With it, it calls search_docs("payin mtn benin") and gets this from the latest snapshot:

### Sandbox environment › Payin (collection)
section: sandbox | source: https://docs.feexpay.me/?section=sandbox&version=v2

| Channel       | Endpoint (POST)                                    | Phone number format          |
| ------------- | -------------------------------------------------- | ---------------------------- |
| MTN Benin     | `/api/transactions/public/requesttopay/mtn`        | any (e.g. `22997000000`)     |
| Moov Benin    | `/api/transactions/public/requesttopay/moov`       | `22901` + 8 digits           |
| Celtiis Benin | `/api/transactions/public/requesttopay/celtiis_bj` | `22901` + 8 digits           |
| Wave CI       | `/api/transactions/public/requesttopay/wave_ci`    | starts with `225`            |
…

Then it writes the integration from that, with the right path, the right body and the right phone format, and cites its source.

Unofficial project. It is not affiliated with or endorsed by FeexPay. The documentation content belongs to FeexPay; the original is at docs.feexpay.me.

Why

FeexPay is a payment aggregator for West and Central Africa (Benin, Côte d'Ivoire, Togo, Senegal, Congo, Burkina Faso): Mobile Money and card collection, payouts to Mobile Money wallets, SDKs and webhooks.

Its documentation is hard for AI tools to read:

• It is a JavaScript app. A plain HTTP fetch (what most assistants use) gets an empty page that just says "Documentation".
• Most of the API reference is hidden behind tabs. The Payin and Payout pages show one country and one network at a time, so even a browser-based scraper that only reads the visible page misses most endpoints.

This server renders every page, clicks through every country and network tab, converts everything to clean Markdown (tables and code blocks included) in English and French, and serves it to your assistant.

Features

• Complete. All sections of API v2, with each country/network variant (endpoint URL, method, request body, headers, code example, response).
• Fresh. A GitHub Action re-scrapes the docs every day and commits the changes. The server reads that snapshot from GitHub at runtime, so you get updates without upgrading the package.
• Works offline. If GitHub is unreachable, the server falls back to the snapshot bundled with it.
• Light and safe. Two runtime dependencies. No browser, no API key, no FeexPay credentials. It only reads documentation: it cannot make payments or call the FeexPay API.

Coverage

All 40 sections of the documentation (API v2), in English and French:

Installation

You need Node.js 18+ and git. The commands below run the server straight from GitHub.

Claude Code

claude mcp add feexpay-docs -- npx -y github:misterkante/feexpay-docs-mcp

Claude Desktop, Cursor, Windsurf

Add this to your MCP config file. That is claude_desktop_config.json for Claude Desktop, ~/.cursor/mcp.json or .cursor/mcp.json for Cursor, and ~/.codeium/windsurf/mcp_config.json for Windsurf:

{
  "mcpServers": {
    "feexpay-docs": {
      "command": "npx",
      "args": ["-y", "github:misterkante/feexpay-docs-mcp"]
    }
  }
}

VS Code (GitHub Copilot)

Put this in .vscode/mcp.json:

{
  "servers": {
    "feexpay-docs": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "github:misterkante/feexpay-docs-mcp"]
    }
  }
}

From a local clone

git clone https://github.com/misterkante/feexpay-docs-mcp.git
cd feexpay-docs-mcp && npm ci --omit=dev
# then point your client at: node /absolute/path/to/feexpay-docs-mcp/src/server.js

Tools

Every section is also exposed as an MCP resource: feexpay-docs://{lang}/{id}, for example feexpay-docs://fr/services-payout.

All tools accept lang: "en" (default) or "fr".

Example prompts

• "Using the FeexPay docs, add Mobile Money payment (MTN and Moov Benin) to my NestJS backend and poll the status until it is final."
• "Which networks can I send payouts to in Côte d'Ivoire, and what body does each endpoint expect?"
• "Integrate the FeexPay React SDK in my checkout page and handle the callback."
• "What does the FeexPay error ERR_IP_NOT_AUTHORIZED mean?"

Configuration

Set these as environment variables, for example in the env field of your MCP client config:

Remote content is cached in memory for 6 hours.

How it works

docs.feexpay.me ──(daily GitHub Action: Playwright + Chromium)──▶ docs/{en,fr}/*.md + manifest.json ──▶ this repo (main)
                                                                                                          │
your assistant ◀──── stdio ──── feexpay-docs-mcp (Node) ◀──── raw.githubusercontent.com (fallback: bundled copy)

1. scraper/scrape.js reads the list of sections from the site's JavaScript bundle. It reads it as text and never runs it in Node.
2. Each section is rendered in a headless Chromium. The scraper clicks every tab and turns each variant into Markdown with its own ## Tab: <country> › <network> heading.
3. If a scrape looks broken (too many failed pages, FeexPay down…), it is discarded and the previous snapshot is kept.
4. docs/manifest.json records a content hash and a last-change date for each page. When nothing changed, the job commits nothing.

The snapshot is plain Markdown. You can also browse it directly in docs/.

Development

npm ci
npm test                                     # unit tests and an end-to-end MCP test over stdio
npx playwright install chromium              # once, or use CHROMIUM_PATH=/usr/bin/chromium
npm run scrape                               # rebuild docs/ from docs.feexpay.me
npm run scrape -- --langs fr --only services-payout   # quick run, output left in docs.tmp-*/

Contributions are welcome. A likely one: the scraper may need adjusting when FeexPay changes its site. In that case the daily job fails and the current snapshot stays in place.

Content and license

• The code in this repository is under the MIT license.
• The documentation content in docs/ is © FeexPay. It is mirrored, unchanged apart from the conversion to Markdown, so developers and AI tools can read it, and every page links to its source. If you represent FeexPay and want it changed or removed, please open an issue.
• The docs are third-party content served to AI assistants as reference data. Always check critical details (amounts, security, production keys) against docs.feexpay.me and your FeexPay dashboard.