Antrophic Model Context Protocol (MCP) integration with SGlang ?

[wugutech]

Do you guys aware about this MCP ?
Could this be valuable to SGLang?

Reference with Ollama:
https://www.shamimbhuiyan.ru/blogs/model-context-protocol-mcp-connecting-local-llms-to-various-data-sources
ollama/ollama#7865

[abzb1]

MCP (Model Context Protocol) is a way to connect local LLMs to external data sources, and conceptually it shares similarities with SGLang’s Function Calling interface.

In particular, SGLang supports function-style interactions via its backend API, which you can refer to here:
https://docs.sglang.ai/backend/function_calling.html#Non-Streaming-Request

That said, there is currently no official mention of direct MCP integration with SGLang. If you're interested in such a feature, it may be worth opening a feature request to explore the potential for support.

Note that MCP currently supports the Anthropic API, whereas SGLang implements an OpenAI-compatible server interface. Therefore, some adaptation or extension may be required to make MCP tools work seamlessly with SGLang's function calling mechanism.

[rektide]

Note that MCP currently supports the Anthropic API, whereas SGLang implements an OpenAI-compatible server interface.

As of May, OpenAI now seemingly supports MCP. I haven't dug in to see exactly what this looks like, but heads-up.
https://community.openai.com/t/introducing-support-for-remote-mcp-servers-image-generation-code-interpreter-and-more-in-the-responses-api/1266973

Being able to bring in an ecosystem of tools is very high on my personal list of wants for SGLang. 🚀

[rektide]

One useful reference during implementation might be Qwen3-Coder's recent qwen3coder_tool_parser.py file:

https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/blob/main/qwen3coder_tool_parser.py