higress-ai-search

The Higress AI-Search MCP Server integrates Higress AI-Search capabilities into your LLM workflows, enabling real-time web, academic, and internal knowledge lookup to augment model responses. It uses Higress plugins to query Google/Bing/Quark for general web results, Arxiv for scientific papers, and internal knowledge bases you specify, then feeds relevant results back to the model to improve accuracy and context. Configure the server with your Higress URL and the target language model, along with a description of internal knowledge bases you want to surface during conversations. With the supplied environment variables, you can tailor the search sources and the model used for response generation. The server is designed to work with uv/uvx tooling for quick setup or with a local development path if you’re cloning the repository.

To use it, run the MCP server via uvx (or uv) and point your AI pipeline to pass the enriched context returned by the Higress AI-Search integration. Ensure Higress is configured with the ai-search and ai-proxy plugins to enable the extensions, and set HIGRESS_URL to the Higress service endpoint. The MODEL environment variable selects the LLM model for generation, while INTERNAL_KNOWLEDGE_BASES lets you describe internal sources that should be prioritized when answering questions.

Prerequisites

• Install uv (and optionally uvx) from the uv project: https://github.com/astral-sh/uv
• Have Higress configured with the ai-search and ai-proxy plugins as described in the Higress docs
• Access to a running Higress instance at the endpoint you configure via HIGRESS_URL

Option A: Install and run via uvx (recommended for quick setup)

1. Install uvx and the mcp server package from PyPI:

# Prerequisites
pipx ensurepath
pipx install uvx  # if you don’t already have uvx, or install via your preferred method

2. Run the MCP server using uvx with the example configuration (adjust env vars as needed):

# Example: start via uvx (uses PyPI package higress-ai-search-mcp-server)
uvx higress-ai-search-mcp-server

3. Verify the server is reachable and listening on the configured Higress endpoint.

Option B: Install and run via uv with local development

1. Clone the repository locally:

git clone https://github.com/your-org/higress-ai-search-mcp-server.git
cd higress-ai-search-mcp-server

2. Start the server in development mode by pointing uv to the source directory:

uv --directory path/to/src/higress-ai-search-mcp-server run higress-ai-search-mcp-server

3. Ensure the environment variables are set (HIGRESS_URL, MODEL, INTERNAL_KNOWLEDGE_BASES) before starting.

Tips and caveats:

• HIGRESS_URL defaults to http://localhost:8080/v1/chat/completions if not provided; override it to point at your Higress service.
• MODEL should be set to the desired language model (e.g., qwen-turbo). Different models may influence latency and result quality.
• INTERNAL_KNOWLEDGE_BASES is a descriptive string; include the names or descriptions of internal docs you want surfaced to improve relevance.
• If using uv with a local repository, ensure the path to the source matches the uv command arguments and that the entry point matches the package name.
• When using the Higress ai-search plugin, confirm the ai-proxy plugin is also enabled to route queries correctly.
• Network access between the MCP server, Higress service, and external search engines is required; ensure firewall rules permit necessary traffic.
• Monitor logs for any authentication or plugin compatibility issues with Higress updates, and keep the HIGRESS_URL and MODEL values in sync with your deployment.