KB-Bridge is a Python-based MCP server that provides intelligent knowledge base search and retrieval with support for multiple backends. It exposes an MCP endpoint that clients can query to perform hybrid, semantic, keyword, and full-text search across configured knowledge sources, then synthesize and refine answers with optional quality checks. The server integrates with backends like Dify and can be extended to additional providers via its retrieval and re-ranking tooling. You can call the MCP tools (notably the default 'assistant') to perform searches and retrieve structured results suitable for downstream agents (e.g., Dify workflows or custom MCP clients). The available tooling includes modules for assistant (answer extraction), file_discover (semantic file discovery), file_lister (dataset inspection), keyword_generator (LLM-based keyword expansion), retriever (various search methods), and file_count (dataset metrics), enabling end-to-end workflow from query understanding to refined answers.

Prerequisites:

• Python 3.8+ installed on the host
• Pip available in PATH

Step 1: Create and activate a virtual environment (recommended)

• python -m venv venv
• source venv/bin/activate # on macOS/Linux
• venv\Scripts\activate # on Windows

Step 2: Install the KB-Bridge package

• pip install kbbridge

Step 3: Prepare environment/configuration

• Create a .env file with your credentials and endpoints as described in the configuration section of the README. Example: RETRIEVAL_ENDPOINT=https://api.dify.ai/v1 RETRIEVAL_API_KEY=your-retrieval-api-key LLM_API_URL=https://your-llm-service.com/v1 LLM_MODEL=gpt-4o LLM_API_TOKEN=your-token-here

  Optional

  RERANK_URL=https://your-rerank-api.com RERANK_MODEL=your-rerank-model

Step 4: Run the server

• python -m kbbridge.server --host 0.0.0.0 --port 5210

Optional: If you prefer Docker for local development, see the deployment notes in the README and use a Dockerfile to build kbbridge:latest and run with environment variables loaded from a .env file.

Tips and caveats:

• Ensure the environment variables for retrieval and LLM services are correctly set; the server will not start correctly without required credentials for your chosen backends.
• The MCP endpoint is exposed at http://<host>:5210/mcp; use this URL from MCP clients or Dify agent workflows.
• If you update any dependencies or switch backends, re-run installation steps or rebuild the Docker image as needed.
• For development and testing, install development dependencies via: pip install -e ".[dev]" and run tests with pytest.
• When using Docker, map port 5210 to the host and provide the same environment variables via --env-file .env for parity with local runs.