Install

Three ways to put closegate in front of your AI agent.

Pick the install path that matches how your team works. All three are free, Apache-2.0, and run on your own infrastructure. 986 passing tests, 19 MCP tools, 11 jurisdictions out of the box.

Path 1 · Python library

Just the policy gate

The chokepoint primitive as a pure Python function. No MCP server, no Docker, no FastAPI. Wrap it around any LLM tool call. Works with Claude Agent SDK, OpenAI Apps SDK, LangGraph, Mastra, or your own custom agent loop.

terminal

pip install closegate-policy

example.py

from decimal import Decimal
from closegate_policy.gate import (
    Action, ActorContext, MatchContext, PolicyConfig, evaluate,
    Allow, RequireHumanApproval, Deny,
)

decision = evaluate(
    action=Action.CONFIRM,
    match=MatchContext(
        match_id="m-1",
        state="PROPOSED_MATCH",
        proposed_by="llm:claude",
        source="llm",
        match_type="exact_match",
        amount_usd=Decimal("15000"),
    ),
    actor=ActorContext(id="engine:autoconfirm", kind="engine"),
    accounts={"5000-COGS"},
    rationale=None,
    config=PolicyConfig(...),
)

match decision:
    case Allow():
        ...  # mutate state, write audit, ship
    case RequireHumanApproval(clause):
        ...  # surface approval envelope with clause.text
    case Deny(clause):
        ...  # raise PolicyViolation

Dependencies: pydantic, py-moneyed, babel, holidays, pyyaml PyPI page →

Path 2 · MCP server

Drop into Claude Desktop, Cursor, OpenAI Apps SDK

The Model Context Protocol server with 19 finance tools tier-routed by reversibility. Speaks stdio, HTTP, and SSE transports — works with every MCP-compliant client.

terminal

pip install closegate-engine
# or run without installing:
uvx closegate-engine mcp serve --transport stdio

Claude Desktop config

~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "closegate": {
      "command": "uvx",
      "args": ["closegate-engine", "mcp", "serve", "--transport", "stdio"],
      "env": {
        "CLOSEGATE_ACTOR_ID": "human:your-name",
        "CLOSEGATE_DB_PATH": "/Users/you/.closegate/recon.db"
      }
    }
  }
}

Cursor

.cursor/mcp.json

{
  "mcpServers": {
    "closegate": {
      "command": "uvx",
      "args": ["closegate-engine", "mcp", "serve"],
      "env": { "CLOSEGATE_ACTOR_ID": "human:your-name" }
    }
  }
}

Smithery (one line)

bash

npx -y @smithery/cli install closegate-engine --client claude

Or run as HTTP server: --transport http --port 8001 PyPI page →

Path 3 · Docker

Self-host the full reference workflow

The consolidated demo image — engine + agent + Vue workspace in one container. Perfect for fly.io, Render, Railway, or your own Kubernetes cluster. Multi-arch (linux/amd64 + linux/arm64) on GHCR.

terminal

docker run --rm \
  -p 8000:8000 -p 8001:8001 -p 5173:5173 \
  -e CLOSEGATE_ACTOR_ID="human:you" \
  ghcr.io/neul-labs/closegate-demo:0.1.0

Or via Docker Compose (three separate services)

bash

git clone https://github.com/neul-labs/closegate
cd closegate
docker compose up
# → http://localhost:5173 — the workspace

Multi-arch images available on ghcr.io/neul-labs/closegate-engine Repo →

Verify the install

One-line check

After install, confirm the gate works with a sanity import:

bash

python -c "from closegate_policy.gate import evaluate, Action; print('closegate-policy ok')"

For the MCP server, run closegate-engine mcp inspect to list all 19 tools and their tier annotations. The output is the canonical surface your AI client will see.

Inbound

Need help wiring closegate into your stack?

Most teams need ~80 engineering hours total to go from clone to first close-cycle pilot. DM the maintainer if you want to compare notes on integration shape — happy to talk regardless of fit.

Open a conversation Read the docs