Feature Request: Support for accessing instructions from InitializeResult

[Mrhy1996]

Feature Request: Support for accessing instructions from InitializeResult

Metadata-Version: 2.4
Name: langchain-mcp-adapters
Version: 0.2.1

TL;DR

Problem: MCP servers return valuable instructions in their InitializeResult, but langchain-mcp-adapters discards this data at line 481 of tools.py.

Solution: Create a new instructions.py module with functions like get_instructions() to access this important metadata.

Impact: Zero breaking changes, follows existing package patterns, enables better MCP server integration.

---

Description

According to the MCP protocol specification, the initialize() method returns an InitializeResult that includes an instructions field in the ServerCapabilities. However, in the current implementation of langchain-mcp-adapters, this result is discarded after calling await tool_session.initialize().

Current Behavior

In tools.py at line 481:

async with create_session(
    connection, mcp_callbacks=mcp_callbacks
) as tool_session:
    await tool_session.initialize()  # ← Result is discarded here
    tools = await _list_all_tools(tool_session)

The InitializeResult returned by initialize() contains valuable information including the instructions field which provides server-specific guidance for AI models, but it's currently being thrown away.

Expected Behavior

It would be beneficial to have a way to access the instructions field from the InitializeResult. This could be useful for:

1. Providing context to LLM agents about how to use the MCP server effectively
2. Displaying server-specific instructions to users
3. Programmatically accessing server metadata and capabilities
4. Following the MCP protocol specification properly

Proposed Solution

I suggest creating a new module file instructions.py in the langchain_mcp_adapters package to handle server capabilities, including the instructions field.

Thank you for maintaining this excellent library! This feature would greatly enhance the usability of MCP servers with LangChain applications.