Ms Agent

ms-agent Skills

This skill connects you to the ms-agent Capability Gateway — a unified interface to ms-agent's projects, components, and atomic tools, exposed as MCP tools.

Setup

Verify ms-agent is installed:

python scripts/check_ms_agent.py

The MCP server must be configured in your agent's config. Pick the section that matches your agent host.

nanobot config.json

{
  "tools": {
    "mcpServers": {
      "ms-agent": {
        "command": "python3",
        "args": ["-m", "ms_agent.capabilities.mcp_server"],
        "env": {"PYTHONPATH": "/path/to/ms-agent"},
        "toolTimeout": 300,
        "enabledTools": ["*"]
      }
    }
  }
}

OpenClaw openclaw.json

{
  "mcp": {
    "servers": {
      "ms-agent": {
        "command": "/absolute/path/to/python",
        "args": ["-m", "ms_agent.capabilities.mcp_server"],
        "env": {"PYTHONPATH": "/path/to/ms-agent"}
      }
    }
  }
}

Hermes Agent config.yaml

mcp_servers:
  ms-agent:
    command: "python3"
    args: ["-m", "ms_agent.capabilities.mcp_server"]
    env:
      PYTHONPATH: "/path/to/ms-agent"

Once configured, all 30 capabilities below are available as MCP tools. The tool prefix depends on your agent host:

Capability Index

Deep Research

Document Research

Financial Research

Web Search

Code Generation

Video Generation

Agent Delegation

Code Validation

File Editing

Quick Decision Guide

User wants to...
│
├── Research a topic in depth (20-60 min)
│   └── submit_research_task → check_research_progress → get_research_report
│
├── Analyze documents or URLs (5-20 min)
│   └── submit_doc_research_task → check_doc_research_progress → get_doc_research_report
│
├── Financial analysis with data & sentiment (20-60 min)
│   └── submit_fin_research_task → check_fin_research_progress → get_fin_research_report
│
├── Search the web for quick info
│   └── web_search(query="...", engine_type="arxiv|exa|serpapi")
│
├── Generate a software project from requirements (10-30 min)
│   └── submit_code_genesis_task → check_code_genesis_progress → get_code_genesis_result
│
├── Generate a short video (~ 20 min)
│   └── submit_video_generation_task → check_video_generation_progress → get_video_generation_result
│
├── Delegate a complex multi-step task to an AI agent
│   ├── Short task (< 3 min) → delegate_task(query="...")
│   └── Long task → submit_agent_task → check_agent_task → get_agent_result
│
├── Validate code for errors
│   ├── Full project → lsp_check_directory(directory="src/", language="typescript")
│   └── Single file → lsp_update_and_check(file_path="...", content="...", language="...")
│
└── Edit a file precisely
    ├── Know exact text → replace_file_contents (concurrent-safe)
    └── Know line numbers → replace_file_lines

Async Pattern

All project-level capabilities (deep research, doc research, financial research, code generation, video generation, agent delegation) support an async submit/check/get pattern. This does NOT block the agent — continue handling other messages while the task runs in the background.

1. submit_*_task(...)      → returns task_id immediately
2. check_*_progress/task(task_id)   → poll status (repeat every few minutes)
3. get_*_result/report(task_id)     → retrieve final result when completed
4. cancel_*_task(task_id)  → cancel if no longer needed (agent delegation only)

Read the reference files for detailed SOP workflows for each capability.