MCP Servers

This folder contains only MCP server infrastructure. Keep application code in its own application repository or folder.

For the automated quant/data analyst system, this repository provides the tool layer:

• mcp-gateway: exposes all local MCP servers through one HTTP/SSE gateway.
• finance-mcp: custom market-data and technical-analysis MCP server.
• tavily: news and sentiment search MCP server.
• sqlite: local persistence MCP server.
• filesystem: report file writing MCP server.
• custom-mcp: general helper MCP server.
• fetch, memory, github: additional MCP services available through the gateway.

The quant application lives separately at:

C:\MachineLearning\Automated_Quant_Data_Analyst

That app calls this MCP gateway programmatically. This separation is intentional:

• MCP servers can be deployed, upgraded, and monitored as independent infrastructure.
• The application can be deployed independently and point at any compatible MCP gateway.
• Runtime tools stay behind a narrow MCP boundary instead of being imported directly into app code.

Runtime Topology

Automated_Quant_Data_Analyst app
        |
        | HTTP/SSE MCP client
        v
http://localhost:8000  or  http://host.docker.internal:8000
        |
        v
mcp-gateway
        |
        +-- finance-mcp      market prices, OHLCV, RSI, SMA, volatility, trend
        +-- tavily           recent market news and sentiment context
        +-- sqlite           local structured persistence
        +-- filesystem       report output under ./data/fs
        +-- custom-mcp       generic custom helpers
        +-- fetch/memory/github

Gateway URLs

When the Docker Compose stack is running, the gateway exposes each server at:

http://localhost:8000/servers/finance/sse
http://localhost:8000/servers/tavily/sse
http://localhost:8000/servers/sqlite/sse
http://localhost:8000/servers/filesystem/sse
http://localhost:8000/servers/custom/sse

Inside another Docker container that is not part of this Compose network, use:

http://host.docker.internal:8000/servers/finance/sse

The gateway configuration is in mcp.json.

Start The MCP Stack

From this folder:

cd C:\MachineLearning\MCP_servers
docker compose up -d --build

Check health:

docker ps
Invoke-RestMethod http://localhost:8000/status

Finance MCP

The finance MCP is documented in detail here:

finance-mcp\README.md

It is registered in:

• docker-compose.yml as service finance-mcp
• mcp.json as gateway server finance

Data Volumes

The current stack uses:

• sqlite-data Docker volume for SQLite MCP database storage.
• memory-data Docker volume for memory MCP state.
• ./data/fs bind mount for filesystem MCP output.

The filesystem MCP only allows writes under /data/fs inside its container. Host-visible files appear under:

C:\MachineLearning\MCP_servers\data\fs

For reports, the quant app writes through the filesystem MCP to:

/data/fs/reports

which appears on the host as:

C:\MachineLearning\MCP_servers\data\fs\reports

Environment

Copy .env.example to .env and set required keys:

MCP_PROXY_VERSION=0.12.0
TAVILY_API_KEY=...
GITHUB_PERSONAL_ACCESS_TOKEN=...

Tavily requires TAVILY_API_KEY. GitHub requires GITHUB_PERSONAL_ACCESS_TOKEN if that service is enabled.

Deployment Boundary

Do not add application code to this folder.

Use this folder for:

• MCP server code
• MCP server Dockerfiles
• MCP gateway configuration
• MCP infrastructure documentation

Use C:\MachineLearning\Automated_Quant_Data_Analyst for:

• FastAPI app code
• application Dockerfile and compose file
• runtime skill files used by the app
• app-specific docs and tests