Using Mcp

Before calling mcp.call(), you MUST first query and confirm through scripts:

1. Whether the tool exists on that server (via get_tools.py) - REQUIRED
2. The tool's parameter definition (via get_tool_schema.py) - REQUIRED
3. Server list (via get_servers.py) - Optional, usually provided in user prompts

STRICTLY FORBIDDEN to call MCP tools based on imagination or experience without querying first. Even if you think a tool "should" exist, you must query to confirm it first.

2. mcp.call() is NOT a tool name

STRICTLY FORBIDDEN to call the following as tool names:

• from sdk.mcp import mcp
• mcp.call
• mcp.call()

mcp.call() is a Python SDK method that must be executed within Python code using the run_skills_snippet tool.

Wrong Example:

# Wrong! Do NOT do this
tool_call(name="from sdk.mcp import mcp", arguments={})
tool_call(name="mcp.call", arguments={...})

Correct Example:

# Correct! Use run_skills_snippet tool
run_skills_snippet(
    python_code="""
from sdk.mcp import mcp
result = mcp.call(...)
"""
)

Key Principles

Script Execution

Scripts use standard command-line arguments, for example:

• python scripts/get_servers.py
• python scripts/get_tools.py --server-name <服务器名称>

In Agent environment, use shell_exec tool to execute scripts:

# 获取服务器列表
shell_exec(
    command="python scripts/get_servers.py"
)

# 获取指定服务器的工具列表
shell_exec(
    command="python scripts/get_tools.py --server-name <服务器名称>"
)

# 获取工具 Schema
shell_exec(
    command="python scripts/get_tool_schema.py --server-name <服务器名称> --tool-name <工具名称>"
)

MCP Tool Calling

Important Note: mcp.call() is NOT a standalone tool, but an SDK method called within Python code.

In Agent environment, use run_skills_snippet tool to execute Python code containing mcp.call() (MUST first query and confirm server and tool existence via scripts):

# 使用 run_skills_snippet 工具执行以下 Python 代码
run_skills_snippet(
    python_code="""
from sdk.mcp import mcp

result = mcp.call(
    server_name="<服务器名称>",  # 必须是通过 get_servers.py 查询到的真实服务器名
    tool_name="<工具名称>",      # 必须是通过 get_tools.py 查询到的真实工具名
    tool_params={"参数": "值"}  # 必须符合 get_tool_schema.py 返回的 Schema
)

if result.ok:
    print(result.content)