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.

[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.

[normand1]

GM! Is langchain-mcp still the package everyone is using for MCP? Planning on integrating this into a tool soon and wondering what the best approach is in working with langchain these days since I noticed langchain-mcp still hasn't been added to the Langchain Package registry yet. I will report back my experience implementing it if still looking for feedback

[krzemian]

From my (very limited and novice) perspective MCP is quickly gaining market adoption and it would seem like a natural step for LangGraph to integrate with it, perhaps eventually even having the tool calling protocol completely handed over to MCP routers at times.
Like I said, this is just a narrow perspective of mine, but so far it seems like it's falling into that place.

All in all, I'm curious about the devs perspective on that one.

EDIT: One POC attempt I saw was https://medium.com/@pranavdhoolia/building-our-own-universal-assistant-with-langgraph-and-mcp-50b503fea1f0 & https://www.youtube.com/watch?v=y6MG-aZqmFw

[IgorGanapolsky]

I'm still confused about the advantages of MCP over other agentic solutions...

[raffareis]

I'm still confused about the advantages of MCP over other agentic solutions...

Well, my understanding is that MCP transfer the responsibility of tool implementation from the dev to the Vendor. It's what the vendor understands an LLM needs to interact with its services.

IMO it makes more sense, the same way we use APIs and not SQL Queries to interact with services on the web.

[IgorGanapolsky]

But can't WindSurf IDE do all of that for us on our machines?

…

On Sun, Mar 2, 2025, 1:36 PM Rafael Reis ***@***.***> wrote: I'm still confused about the advantages of MCP over other agentic solutions... Well, my understanding is that MCP transfer the responsibility of tool implementation from the dev to the Vendor. It's what the vendor understands an LLM needs to interact with its services. IMO it makes more sense, the same way we use APIs and not SQL Queries to interact with services on the web. — Reply to this email directly, view it on GitHub <#28352 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AABRD6L7AH2IFI5AXJPVA632SNFSZAVCNFSM6AAAAABSQIN2LKVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEMZWG43DIOA> . You are receiving this because you are subscribed to this thread.Message ID: <langchain-ai/langchain/repo-discussions/28352/comments/12367648@ github.com>

[vbarda]

@cemkod check out https://github.com/langchain-ai/langchain-mcp-adapters