feat(mcp): add mcp client and server

[paul-nechifor]

Problem

We need to move towards greater support for MCP.

Closes DIM-533

https://linear.app/dimensional/issue/DIM-533/add-mcp-client-and-server

Solution

• Moving towards deprecating Agent. It has been efectivelly split between McpClient and McpServer.
• McpServer exposes all the @skill functions as MCP tools.
• McpClient starts a langchain agent which coordinates calling tools from McpServer.
• McpClient is not strictly necessary. You can register the MCP server in claude code and make it call tools.
• McpClient also listens to /human_input so it can be used through the humancli like Agent can.

Next tasks:

• Adding an easy way of exposing MCP tasks.

Breaking Changes

None

How to Test

Run

uv run dimos run unitree-go2-agentic-mcp

...instead of unitree-go2-agentic and interact with it the same way through humancli: start exploring, stop, move forward 2 meters, etc.

[greptile-apps]

Greptile Summary

Refactored MCP implementation from TCP-based stdio bridge to direct HTTP-based FastAPI server/client architecture. The new approach simplifies the integration with Claude Code by removing the intermediate bridge layer and using HTTP JSON-RPC directly.

Major Changes:

• Moved MCP implementation from dimos/protocol/mcp/ to dimos/agents/mcp/
• Replaced TCP socket server with FastAPI HTTP server (McpServer) on port 9990
• Created HTTP-based MCP client (McpClient) using httpx for JSON-RPC communication
• Removed dependency on mcp>=1.0.0 package, implementing custom JSON-RPC handlers
• Added comprehensive test coverage (integration and unit tests) with fixture-based testing
• Updated unitree_go2_agentic_mcp blueprint to use new client/server modules

Issues Found:

• Critical: test_mcp_client.py:62 returns ValueError instead of raising it, breaking error handling logic
• Critical: test_mcp_server.py:105 assertion will fail due to mismatched error response format
• Minor: Grammar error in mcp_client.py:150 ("You will updated")
• Minor: Typo in README.md:25 ("youse")

Confidence Score: 3/5

• PR has solid architecture but contains critical test bugs that will cause runtime failures
• The refactoring from TCP to HTTP is well-designed and the core implementation is sound. However, there are two critical bugs in test code: ValueError being returned instead of raised (line 62 in test_mcp_client.py) and a failing assertion (line 105 in test_mcp_server.py). These will cause test failures and indicate the tests haven't been fully validated. The grammar/typo errors are minor but should be fixed.
• Pay close attention to dimos/agents/mcp/test_mcp_client.py and dimos/agents/mcp/test_mcp_server.py for the critical logic errors

Important Files Changed

Filename	Overview
dimos/agents/mcp/README.md	comprehensive documentation for MCP server setup and usage, contains typo on line 25
dimos/agents/mcp/mcp_client.py	MCP client implementation with HTTP-based JSON-RPC communication to MCP server, contains grammar error on line 150
dimos/agents/mcp/mcp_server.py	FastAPI-based MCP server exposing robot skills as MCP tools, well-structured with proper error handling
dimos/agents/mcp/test_mcp_client.py	integration tests for MCP client, contains critical bug on line 62 where ValueError is returned instead of raised
dimos/agents/mcp/test_mcp_server.py	unit tests for MCP server request handling, contains assertion issue on line 105
dimos/robot/unitree/go2/blueprints/agentic/unitree_go2_agentic_mcp.py	updated to use new MCP client/server architecture with autoconnect

Sequence Diagram

sequenceDiagram
    participant Claude as Claude Code
    participant Client as McpClient
    participant Server as McpServer
    participant Skills as Robot Skills

    Note over Claude,Skills: Initialization Phase
    Claude->>Server: HTTP POST /mcp (initialize)
    Server-->>Claude: Server capabilities & info
    Claude->>Server: HTTP POST /mcp (tools/list)
    Server->>Skills: get_skills()
    Skills-->>Server: List of SkillInfo
    Server-->>Claude: Available tools with schemas

    Note over Claude,Skills: Tool Execution Phase
    Claude->>Client: Human input message
    Client->>Client: Queue message
    Client->>Client: LangChain agent processes
    Client->>Server: HTTP POST /mcp (tools/call)
    Server->>Skills: RPC call to skill
    Skills-->>Server: Result or None
    Server-->>Client: JSON-RPC result
    Client->>Client: Append to history
    Client->>Claude: Publish agent message

    Note over Client,Skills: Image Handling
    Skills-->>Server: Result with agent_encode()
    Server-->>Client: Content with image data
    Client->>Client: Add UUID, append to history
    Client->>Claude: Tool response + image message

Loading

_{Last reviewed commit: 2fdf0cb}

[greptile-apps]

_{27 files reviewed, 4 comments}

_{Edit Code Review Agent Settings | Greptile}

[spomichter]

@paul-nechifor should we just fork openclaw? Theyve build a lot of robust MCP and hosted gateway infra, runs on all architectures, etc

[paul-nechifor]

@paul-nechifor should we just fork openclaw? Theyve build a lot of robust MCP and hosted gateway infra, runs on all architectures, etc

I don't think we need to fork openclaw, we just want to connect it to our MCP server. And we'll want to publish a SKILL.md file to ClawHub so openclaw knows how to use our server.