Suggest making the API into an MCP server

[smileyboy2019]

Do you need to file a feature request?

• [ ] I have searched the existing feature request and this feature request is not already filed.
• [ ] I believe this is a legitimate feature request, not just a question or bug.

Feature Request Description

The service protocol of MCP supports multiple interfaces and can open up services for clients to implement graph writing and query services.

Additional Context

No response

[danielaskdd]

There are several points that require clarification:

• What functionalities need to be integrated into the MCP server?
• What is the appropriate approach for introducing the MCP server to the LLM, considering that each application integrates the LightRAG server for different purposes?

[zhuyanhuazhuyanhua]

MCP 服务器（Multi - Client Proxy Server，多客户端代理服务器）：

Connection management: managing connections to multiple clients (like LLM, LightRAG servers, etc.), including connection establishment, disconnection, and heartbeat detection. Request routing: routing requests to the corresponding service components based on the type and target of the client request, such as directing different query requests to different storage modules. Load balancing: distributing requests across multiple backend service instances to prevent any single service from becoming overloaded, thereby improving overall system performance and availability. Security authentication: verify the identity and authorization of client requests to ensure system security and prevent unauthorized access. Data caching: cache frequently accessed data to reduce backend service pressure and speed up response time. Monitoring and logging: record system to facilitate subsequent problem troubleshooting and performance optimization.

[BireleyX]

I don't think converting the existing API to MCP server is necessary or even the correct approach. MCP servers are API wrappers... to make the interaction with AI consistent. I was also thinking about using MCP with Lightrag for my project, luckily I already found an existing project: https://github.com/shemhamforash23/lightrag-mcp

@smileyboy2019 I haven't tried the MCP server for Lightrag project yet, but it seems to already provide all the necessary Tools to interface with the current functions of LightRag. It just needs a running instance of LightRag API.

[choizhang]

LLM in natural language dialogue currently cannot determine whether a problem requires a local knowledge base, a direct connection to LLM, or online search through intent recognition. Server memory is identified through Custom Instructions starting with 'please remember' and is not accurate, while mem0 mcp mainly focuses on programming. At present, Retrieval in webui is mainly in the inference stage, and the construction is mainly focused on documents. An interface for directly transmitting text can be added