Skip to content

MCP_INTEGRATION.md

Latest commit

 

History

History
72 lines (54 loc) · 1.58 KB

MCP_INTEGRATION.md

File metadata and controls

72 lines (54 loc) · 1.58 KB

MCP Integration

This project now includes a separate MCP adapter server in mcp_server.py.

The adapter does not replace existing logic. It calls your existing FastAPI backend (unimain.py) over HTTP and exposes MCP tools on top.

1) Start Unified Backend

cd SearchEmbedSDK
uvicorn unimain:app --host 127.0.0.1 --port 8000

2) Start MCP Server (stdio)

cd SearchEmbedSDK
python mcp_server.py

Environment variables:

  • CONTEXTCORE_API_BASE_URL (default: http://127.0.0.1:8000)
  • CONTEXTCORE_MCP_TIMEOUT_SECONDS (default: 120)
  • MCP_TRANSPORT (default: stdio)

3) Claude Desktop Config

Add an MCP server entry pointing to mcp_server.py.

{
  "mcpServers": {
    "contextcore-unified": {
      "command": "C:\\Users\\USER\\Documents\\SDKSearchImplementation\\SearchEmbedSDK\\.venv\\Scripts\\python.exe",
      "args": [
        "C:\\Users\\USER\\Documents\\SDKSearchImplementation\\SearchEmbedSDK\\mcp_server.py"
      ],
      "env": {
        "CONTEXTCORE_API_BASE_URL": "http://127.0.0.1:8000"
      }
    }
  }
}

4) Tools Exposed by MCP Server

  • server_info
  • health
  • unified_search
  • run_llm
  • index_scan
  • image_index_status
  • recent_activity
  • file_preflight
  • storage_status

5) Other LLM Interfaces

Any interface that supports MCP stdio can run the same command setup.

For MCP hosts that support non-stdio transports, you can set:

$env:MCP_TRANSPORT="streamable-http"
python mcp_server.py

mcp_server.py is intentionally isolated so backend behavior in unimain.py stays unchanged.