Skip to content

interop.md

Latest commit

 

History

History
61 lines (43 loc) · 1.89 KB

interop.md

File metadata and controls

61 lines (43 loc) · 1.89 KB
Raw

Interop

SkillEngine speaks three agent ecosystems out of the box.

MCP (Model Context Protocol)

As a client — call any MCP server's tools as if they were local:

from skillengine import MCPServerSpec
pool = await runner.connect_mcp_servers([
    MCPServerSpec(command="npx", args=["-y", "@modelcontextprotocol/server-everything"]),
])
# Tools appear in the dispatcher as "<name>__<tool>".

As a server — expose your skills/tools to Claude Desktop / Cursor / Cline:

python -m skillengine.mcp --skill-dir ./skills --name my-skills

The CLI speaks JSON-RPC 2.0 over stdio. Skills become skill__<name> tools.

Spec URIs:

Form Example
Shell mcp+stdio:npx -y @mcp/server-everything
URL mcp+stdio://node?args=server.js&env=DEBUG=1
JSON mcp+command:{"command":"node","args":["s.js"]}

Use skillengine.mcp.parse_mcp_uri() to convert any of these into an MCPServerSpec.

A2A (Agent-to-Agent)

The skillengine.a2a package implements the A2A draft protocol with:

  • AgentCard derived automatically from your skills
  • A2AClient for sending tasks to remote agents
  • AgentRegistry + AgentDiscovery for service discovery
  • PerformanceRouter for picking the fastest healthy agent

Handoffs (OpenAI Agents SDK / Anthropic A2A compat)

A handoff is a synthetic transfer_to_<agent> tool that delegates to another agent. SkillEngine ships factories for the three common targets:

from skillengine import handoff, callable_handoff, agent_handoff, a2a_handoff

h1 = callable_handoff(my_fn, name="formatter")
h2 = agent_handoff(child_runner, name="researcher")
h3 = a2a_handoff(client, endpoint="http://remote", skill_name="analyze",
                 name="analyzer")
runner.add_handoffs([h1, h2, h3])

An optional input_filter lets you trim the message history before transfer, matching the OpenAI Agents SDK extension point.