brvm-mcp

🇫🇷 Version française

MCP server giving AI assistants access to BRVM public data — the West African regional stock exchange serving 8 UEMOA countries (Benin, Burkina Faso, Côte d'Ivoire, Guinea-Bissau, Mali, Niger, Senegal, Togo).

The BRVM publishes no public API. This server aggregates 4 data sources into 18 MCP tools so any AI assistant (Claude Desktop, Claude Code, etc.) can query real-time market data, fundamentals, dividends, volumes, and sector indices.

Disclaimer: For information and analysis only. Does not constitute investment advice (regulated activity under AMF-UMOA). Past performance does not guarantee future results.

18 tools

Official data — brvm.org

Supplementary sources

Related MCP server: yfinance

Quick start

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "brvm": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/sotechdi/brvm-mcp", "brvm-mcp"]
    }
  }
}

Restart Claude Desktop. Then ask: "What are the BRVM stocks with the highest dividend yield?"

Config file locations:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: %APPDATA%\Claude\claude_desktop_config.json

• Windows (Store): %LOCALAPPDATA%\Packages\Claude_*\LocalCache\Roaming\Claude\claude_desktop_config.json

Claude Code

claude mcp add brvm -- uvx --from git+https://github.com/sotechdi/brvm-mcp brvm-mcp

Manual install (pip)

git clone https://github.com/sotechdi/brvm-mcp.git
cd brvm-mcp
python -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate
pip install -r requirements.txt
python server.py

HTTP / Docker mode

For network deployment or integration with non-stdio MCP clients:

docker compose up -d

Exposes a streamable-http MCP endpoint on port 8000. Set MCP_TRANSPORT=sse for legacy SSE clients.

Historical data

brvm.org only exposes the current session. To build trend analysis, run snapshot.py daily after market close (BRVM fixing ~10:45 GMT):

python snapshot.py          # capture today's session (idempotent)
python snapshot.py --stats  # show database coverage

Automate with cron (weekdays at 12:00 GMT):

0 12 * * 1-5  cd /path/to/brvm-mcp && .venv/bin/python snapshot.py >> snapshot.log 2>&1

LLM agent & regulatory compliance

Beyond raw data access, the repository ships a ReAct agent (LangGraph) that consumes the 18 MCP tools and answers market questions in plain language.

Compliance by design. Publishing investment recommendations in the UEMOA zone is a regulated activity requiring a CIB licence from the AMF-UMOA. Rather than relying on prompt instructions alone, every answer passes through an editorial linter that blocks prescriptive vocabulary — buy, sell, we recommend, price target, undervalued — before it is returned. Output that fails the check raises EditorialViolation and is rejected, not silently rewritten: rewriting would hide the drift instead of surfacing it.

The linter matches prescriptive constructions, not isolated words. "The company sold its subsidiary" is a corporate event and passes; "sell SONATEL" does not. Matching is also accent-insensitive, so a missing diacritic cannot slip prescriptive French past the guardrail. Both distinctions are what keep it usable on real bulletins.

pytest tests/test_editorial.py   # 23 prescriptive samples blocked, 14 factual ones passed

Answers therefore stay within factual reporting: prices, yields, volumes, corporate events. No advice, no forecasts.

Architecture

brvm-mcp/
├── server.py               # FastMCP server — 18 tools, stdio/HTTP/SSE transports
├── snapshot.py             # Daily snapshot for historical database
├── brvm_scraper/
│   ├── client.py           # HTTP session, TTL cache (15 min), retry/backoff
│   ├── quotes.py           # Prices, indices, market activity (brvm.org)
│   ├── companies.py        # Listed companies with country filter
│   ├── dividends.py        # Dividends + yield calculation
│   ├── afx_kwayisi.py      # Volumes, fundamentals, sector indices (AFX)
│   ├── richbourse.py       # Previous close, market cap (Rich Bourse)
│   ├── sikafinance.py      # OHLC, dividends (Sika Finance)
│   ├── storage.py          # SQLite historization (UPSERT, stats, performance)
│   └── fundamentals.py     # P/E / EPS extraction from PDF (pdfplumber)
├── agent/
│   ├── graph.py            # LangGraph ReAct agent over the MCP tools
│   ├── tools.py            # LangChain wrappers — all 18 tools
│   ├── prompts.py          # System prompt — analytical role, never advisory
│   ├── editorial.py        # Prescriptive-vocabulary blocker (regulatory guardrail)
│   └── cli.py              # Interactive CLI
└── tests/
    ├── fixture_home.html   # Captured brvm.org HTML for offline tests
    ├── test_parsers.py     # Quote / dividend parsing
    ├── test_storage.py     # SQLite storage
    ├── test_agent_smoke.py # Agent smoke test
    └── test_editorial.py   # Editorial linter — blocking + false-positive guard

Technical notes:

• Regex on page text, not CSS selectors — more resilient to theme changes

• 15-minute TTL cache — BRVM runs a single daily fixing (~10:45 GMT), no need to hammer sources

• Identifiable User-Agent + exponential backoff — respectful of public infrastructure

Pricing

→ sotechdi.github.io/brvm-mcp

Payment: Orange Money, Moov, PayPal, bank transfer — contact@sotechdi.com

Community

Questions, feedback, or ideas? Open a GitHub Discussion — or join the WhatsApp group for French-speaking users (Burkina Faso, Côte d'Ivoire, Sénégal…): [coming soon]

If this server is useful to you, a ⭐ on GitHub helps others find it — thank you!

License

MIT — © 2026 Christian Dondire