FinMCP

FinMCP provides a unified, finance-focused MCP (Model Context Protocol) ecosystem. It weaves together core MCP framework components with specialized modules for deep research, market data access, and trading integrations. Use the configuration in mcp.json to launch multiple servers in parallel, enabling AI-driven research pipelines, real-time data retrieval from Yahoo Finance and Indian markets (NSE/BSE), and direct trading capabilities via Upstox. The Deep Research module powers iterative research with web scraping and discovery, while the Digital Persona and fine-tuning pipelines support document analysis and model refinement for financial contexts. Tools exposed include a REST/streaming interface for research prompts, a market data surface for prices and indicators, and trading endpoints to fetch portfolio data and order information.

Prerequisites:

• Python 3.8+ and pip
• Node.js 16+ with npm
• Git with submodule support
• Access to required API keys (NVIDIA/OpenAI/Firecrawl/Upstox as needed)

Installation steps:

1. Clone repository with submodules:

git clone --recursive https://github.com/Finance-LLMs/FinMCP.git
cd FinMCP

2. Initialize and update submodules to latest:

git submodule update --init --remote --merge --recursive

3. Install core dependencies for each module as needed. Examples:

• Deep Research Engine:

cd Deep-Research
pip install -r requirements.txt

• Upstox/MCP server (Node.js):

cd Upstox-MCP-Server
npm install

• Yahoo Finance/Market modules may have Python requirements as well:

cd Yahoo-Finance
pip install -r requirements.txt

4. Prepare environment variables and configuration files as described in the README (env.local or .env.local files per module).

5. Start the MCP servers using the configuration in mcp.json. For example:

# From project root, ensure you have the required environment and dependencies
# Start all configured MCP servers (the exact command may depend on your setup, e.g., uv or npx integration)
# Example using the mcp.json mapping provided:
# You may use a script or a supervisor to launch the servers listed under mcpServers

6. Verify endpoints and begin interacting with the MCP ecosystem through the exposed API surfaces (research prompts, market data access, and trading endpoints).

Environment variables and API keys placement:

• Deep Research: FIRECRAWL_KEY, NVIDIA_API_KEY, OPENAI_KEY
• Digital Persona: NVIDIA_API_KEY, HF_TOKEN
• Upstox: UPSTOX_CLIENT_ID, UPSTOX_CLIENT_SECRET If running in a containerized or isolated environment, ensure these keys are provided via env files or secure secret management. Common issues include missing API keys, network access restrictions to data sources, and version mismatches across modules. The mcp.json example shows uv and npx usage for launching submodules; adapt paths if your directory structure differs. For local development, you can start with the provided configuration and iterate by enabling/disabling modules in mcp.json as needed. For debugging, check each module's logs for startup errors, port binding, and dependency problems.