nelsonjingusc/goose-mcp-escrow-server

Goose MCP Escrow Server

A minimal MCP server that provides escrow style intent verification, deterministic planning, and reversible execution for Goose agents. The server separates intent parsing, verification, planning, and commit or rollback actions in a clear and auditable workflow.

Features

• Structured intent schema
• Deterministic plan generation
• Lightweight verification rules
• Commit and rollback execution
• Zero external dependencies
• Native MCP tool definitions

Repository Structure

goose-mcp-escrow-server/
  server.py
  mcp.json
  requirements.txt
  README.md
  escrow/
    __init__.py
    intent.py
    verify.py
    planner.py
    executor.py
  examples/
    request_plan.json
    request_execute.json

MCP Tools

escrow_plan

Parses user intent, applies guardrails, and returns a deterministic plan.

Input

{
  "action": "allocate",
  "target": "resource_1",
  "amount": 100,
  "risk_level": "low"
}

Output

{
  "plan_id": "uuid",
  "intent": { ... },
  "constraints": { "risk_level": "low", "max_retries": 1 },
  "steps": [ ... ],
  "explain": "Planned allocate toward resource_1 with risk level low"
}

escrow_execute

Confirms or rolls back a previously generated plan.

Input

{
  "plan_id": "uuid",
  "confirm": true
}

Output

{
  "plan_id": "uuid",
  "status": "committed"
}

Quick Test

1. Generate a plan

python server.py examples/request_plan.json

2. Edit examples/request_execute.json

Replace plan_id with the returned value.

3. Execute the plan

python server.py examples/request_execute.json

Design

• User intent is normalized before any logic is applied.
• Verification applies simple safety rules.
• Planning is deterministic so results can be reproduced.
• Execution supports commit or rollback.

License

MIT License.