Model Context Protocol implementation

[cemkod]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

Support for interacting with MCP servers.

The Model Context Protocol is an open standard that enables developers to build secure, two-way connections between their data sources and AI-powered tools. The architecture is straightforward: developers can either expose their data through MCP servers or build AI applications (MCP clients) that connect to these servers.

https://www.anthropic.com/news/model-context-protocol
https://modelcontextprotocol.io/introduction

Motivation

It would be very useful if we could connect our chains to MCP servers. There are already some useful tool implementations available in MCP server format.
https://github.com/modelcontextprotocol/servers

Proposal (If applicable)

No response

[sheikhuzairhussain]

@baskaryan @hwchase17 @agola11
Is there someone working on this already? If not, I'd love to contribute. Besides, I work at Memorang and I think we might end up using MCP for some of our LangChain/LangGraph workflows anyway.

[johannhartmann]

There is an implementation in https://github.com/rectalogic/langchain-mcp, however it would make much more sense as a toolkit in langchain.

[qingwoo]

i think so

[ccurme]

Maintaining a separate package (e.g., langchain-mcp) has a few advantages:

• We can version it properly, whereas in langchain-community breaking changes in individual integrations are not signaled with a major or minor version bump;
• We can explicitly manage a dependency with mcp, instead of generating ImportErrors and asking users to pip install;
• The package can be more thoroughly integration tested.

The above langchain-mcp has not been added to the LangChain package registry or the docs. Wondering if any of you have concerns about using that package or feedback on its implementation.