k-l-lambda/other-agent-mcp

Other MCP

An MCP (Model Context Protocol) server that provides LangChain.js-powered subagent tools with session management. It allows Claude Code (or any MCP client) to delegate tasks to another LLM while maintaining conversation context.

Features

• Multi-provider support: OpenAI-compatible APIs (Qwen, DeepSeek, local models) and Anthropic-compatible APIs
• Model registry: Configure multiple models and switch between them at runtime
• Session management: Create persistent conversations that maintain context across multiple messages
• Session logging: All conversations are automatically saved as markdown files in the sessions/ directory
• Built-in tools: Agents can read files, list directories, search with grep, and more
• Simple integration: Works with Claude Code via claude mcp add

Installation

cd /path/to/other-mcp
npm install
npm run build

Configuration

Environment Variables

Using .env File

Create a .env file in the project root:

# Provider: 'openai' or 'anthropic'
SUBAGENT_PROVIDER=openai

# API endpoint (for OpenAI-compatible APIs)
SUBAGENT_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1

# Your API key
SUBAGENT_API_KEY=sk-your-api-key-here

# Default model
SUBAGENT_MODEL=qwen-max

# Model registry (JSON format)
# Simple format: model name -> model ID
SUBAGENT_MODELS={"qwen-max":"qwen-max","qwen-turbo":"qwen-turbo","qwen-plus":"qwen-plus"}

Model Registry Format

The SUBAGENT_MODELS environment variable supports two formats:

Simple format - Just map names to model IDs (uses default provider settings):

{
  "qwen": "qwen-max",
  "gpt4o": "gpt-4o-mini",
  "turbo": "qwen-turbo"
}

Full config format - Override provider/baseUrl per model:

{
  "qwen": {
    "provider": "openai",
    "modelId": "qwen-max",
    "baseUrl": "https://dashscope.aliyuncs.com/compatible-mode/v1",
    "apiKey": "sk-qwen-key"
  },
  "claude": {
    "provider": "anthropic",
    "modelId": "claude-3-5-haiku-latest",
    "baseUrl": "https://api.anthropic.com"
  },
  "local": {
    "provider": "openai",
    "modelId": "llama3",
    "baseUrl": "http://localhost:11434/v1",
    "apiKey": "ollama"
  }
}

Adding to Claude Code

Basic Setup (Single Model)

claude mcp add -s user subagent \
  -e "SUBAGENT_PROVIDER=openai" \
  -e "SUBAGENT_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1" \
  -e "SUBAGENT_API_KEY=your-api-key" \
  -e "SUBAGENT_MODEL=qwen-max" \
  -- node /path/to/other-mcp/dist/index.js

Multi-Model Setup

claude mcp add -s user subagent \
  -e "SUBAGENT_PROVIDER=openai" \
  -e "SUBAGENT_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1" \
  -e "SUBAGENT_API_KEY=your-api-key" \
  -e "SUBAGENT_MODEL=qwen-max" \
  -e 'SUBAGENT_MODELS={"qwen-max":"qwen-max","qwen-turbo":"qwen-turbo","qwen-plus":"qwen-plus"}' \
  -- node /path/to/other-mcp/dist/index.js

Multi-Provider Setup

For using multiple providers (e.g., both Qwen and local Ollama):

claude mcp add -s user subagent \
  -e "SUBAGENT_PROVIDER=openai" \
  -e "SUBAGENT_API_KEY=your-qwen-key" \
  -e 'SUBAGENT_MODELS={"qwen":{"provider":"openai","modelId":"qwen-max","baseUrl":"https://dashscope.aliyuncs.com/compatible-mode/v1"},"local":{"provider":"openai","modelId":"llama3","baseUrl":"http://localhost:11434/v1","apiKey":"ollama"}}' \
  -- node /path/to/other-mcp/dist/index.js

MCP Tools

run_agent

Run a LangChain agent with the given prompt (stateless, single-turn).

Parameters:

Example:

mcp__other__run_agent(prompt="Summarize the key points of quantum computing")
mcp__other__run_agent(prompt="Translate to Chinese", model="qwen-turbo")

list_models

List all available models configured in SUBAGENT_MODELS.

Example output:

## Available Models

| Name | Provider | Model ID | Base URL |
|------|----------|----------|----------|
| _default | openai | qwen-max | https://dashscope.aliyuncs.com/compati... |
| qwen-max | openai | qwen-max | (default) |
| qwen-turbo | openai | qwen-turbo | (default) |
| qwen-plus | openai | qwen-plus | (default) |

Use the `model` parameter in `run_agent` to select a model by name.

create_session

Create a new conversation session that maintains context across multiple messages.

Parameters:

Example:

mcp__other__create_session(model="gpt-5.1", system_prompt="You are a helpful assistant")
# Returns: session_id

send_message

Send a message to an existing session and get a response. The conversation history is preserved.

Parameters:

Example:

mcp__other__send_message(session_id="abc-123", message="Hello, how are you?")
mcp__other__send_message(session_id="abc-123", message="What did I just say?")
# The agent remembers the previous message

list_sessions

List all active conversation sessions.

Example output:

## Active Sessions

| Session ID | Model | Messages | Last Active |
|------------|-------|----------|-------------|
| `d7fbfd49...` | gpt-5.1 | 7 | 2025-12-10T07:59:32.956Z |

Total: 1 session(s)

get_session_history

Get the conversation history of a specific session.

Parameters:

delete_session

Delete a conversation session.

Parameters:

Agent Tools (Built-in)

The agent has access to the following read-only tools for exploring the file system:

read_file

Read the contents of a file.

Parameters:

list_directory

List the contents of a directory with file sizes and modification times.

Parameters:

grep

Search for a pattern in files using regular expressions.

Parameters:

glob

Find files matching a glob pattern.

Parameters:

file_info

Get detailed information about a file or directory.

Parameters:

Provider Examples

Qwen (Alibaba Cloud)

SUBAGENT_PROVIDER=openai
SUBAGENT_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
SUBAGENT_API_KEY=sk-your-qwen-key
SUBAGENT_MODEL=qwen-max

DeepSeek

SUBAGENT_PROVIDER=openai
SUBAGENT_BASE_URL=https://api.deepseek.com/v1
SUBAGENT_API_KEY=sk-your-deepseek-key
SUBAGENT_MODEL=deepseek-chat

OpenAI

SUBAGENT_PROVIDER=openai
# SUBAGENT_BASE_URL not needed for official OpenAI
SUBAGENT_API_KEY=sk-your-openai-key
SUBAGENT_MODEL=gpt-4o-mini

Anthropic

SUBAGENT_PROVIDER=anthropic
# SUBAGENT_BASE_URL not needed for official Anthropic
SUBAGENT_API_KEY=sk-ant-your-key
SUBAGENT_MODEL=claude-3-5-haiku-latest

Ollama (Local)

SUBAGENT_PROVIDER=openai
SUBAGENT_BASE_URL=http://localhost:11434/v1
SUBAGENT_API_KEY=ollama
SUBAGENT_MODEL=llama3

Development

# Build
npm run build

# Test tools/list
echo '{"jsonrpc":"2.0","method":"tools/list","id":1}' | node dist/index.js

# Test list_models with config
export SUBAGENT_API_KEY=test
export SUBAGENT_MODELS='{"model1":"gpt-4o","model2":"qwen-max"}'
echo '{"jsonrpc":"2.0","method":"tools/call","params":{"name":"list_models","arguments":{}},"id":2}' | node dist/index.js

License

MIT