@executioncontrolprotocol/mcp

MCP (Model Context Protocol) adapter that exposes an initialized ECP environment to agents.

Status

MCP tools cover discovery, validation, workflow run, and encode/decode. Capability invoke is not yet an MCP tool. To call a single capability outside a workflow, use:

• Fluent API: ecp.invoke(capabilityId).with(input).process()
• CLI: ecp invoke <capability-id> --env environment.ts --input input.json
• HTTP: ecp serve --env environment.ts then POST /v1/invoke (or ecp up for the Ollama demo bridge)

Tools (current)

| Tool | Maps to | | ---- | ------- | | ecp.describe_environment | ecp.describe | | ecp.search | ecp.search | | ecp.validate_workflow | ecp.validate | | ecp.run_workflow | ecp.run | | ecp.encode | ecp.encode | | ecp.decode | ecp.decode | | ecp.get_run_status | in-process run store |

Resources

| Resource | URI | | -------- | --- | | Environment descriptor | ecp://environment/describe | | Capabilities | ecp://capabilities | | Capability detail | ecp://capabilities/{id} | | Policies | ecp://policies | | Run detail | ecp://runs/{runId} |

Usage

import { createEcpMcpServer, serveStdio } from "@executioncontrolprotocol/mcp"
import { environment } from "@executioncontrolprotocol/node"

const env = await environment("demo")
const ecp = await env.init()
const server = createEcpMcpServer({ ecp })
await serveStdio({ ecp })

See AGENTS.md and ecp-overhaul.md §13 for architecture.