Africa MCP — pan-African data for AI agents

A Braynex Services Ltd product. Author: Samuel Orie. License: MIT.

A neutral aggregation + normalization layer that exposes pan-African market data to AI agents (Claude, Cursor, custom agents) as clean MCP tools — behind one typed provider seam, each data source family its own server ("lane"). Sibling to nigeria-mcp, which stays Nigeria-only.

Status: 1 lane shipped (FX & Parallel Rates). npm run verify → build + smoke, fully offline on mocks.

Lanes

v1 banks: Nigeria (CBN), Ghana (BoG), Kenya (CBK), South Africa (SARB), Tanzania (BoT), Rwanda (BNR), Egypt (CBE — scaffolded, TBD).

Related MCP server: latamfx-mcp

Quickstart

npm install
npm run build      # builds shared/ core first, then every lane (npm workspaces)
npm run verify     # build + all smokes — fully offline, zero signup, zero network

Every lane defaults to a mock provider with deterministic fixtures, so the whole repo runs with no signup. Live providers are opt-in via env (see .env.example).

Use the FX lane in Claude Desktop / Code

{
  "mcpServers": {
    "africa-fx-rates": {
      "command": "npx",
      "args": ["-y", "@braynexservices/africa-mcp-fx-rates"],
      "env": { "FX_PROVIDER": "live" }     // live central-bank sources, no key
    }
  }
}

Then ask: "What's the official CBN USD rate and the parallel rate for Nigeria today, and do they reconcile?"

Architecture

Tools never call a source directly — selecting a provider is one env var; adding a source is one file. The moat is maintained cross-provider reconciliation, not aggregation: every official rate is cross-checked against an independent second source, parallel rates are layered on top and never silently blended with official ones, and stale sources degrade to last-known-good with an explicit flag instead of erroring.