cursor

Signetry for Cursor

Two ways to govern coding-agent changes in Cursor with signetry-core.

Quickstart (60 seconds)

# 1. install the kernel (BUSL-1.1 engine; installed from a git tag, not PyPI)
pip install "signetry-core @ git+https://github.com/Signetry/core@v0.8.0"

# 2. scaffold a contract in your repo
cd /path/to/your/repo
signetry init                 # writes a conservative .signetry/admission.yaml

3. Wire Cursor to Signetry

mkdir -p .cursor/rules
cp mcp.json .cursor/mcp.json          # or merge into an existing .cursor/mcp.json
cp signetry.mdc .cursor/rules/        # project rule (advisory)

Reload Cursor; the agent can then call signetry_admit, signetry_verify and signetry_provenance.

Verify it works

# a path outside the contract's allowed_paths must be denied (exit 1)
signetry guard --repo . --path .github/workflows/release.yml; echo "exit=$?"

# a path inside it must be allowed (exit 0)
signetry guard --repo . --path src/app.py; echo "exit=$?"

Cursor has no deterministic pre-write hook, so the enforced gate is CI — see §2 and the note at the end.

1. MCP server (recommended)

Cursor speaks MCP. Add Signetry's server so the agent can run admission / verify / provenance itself. Copy mcp.json to .cursor/mcp.json in your project (or merge into your existing one):

{
  "mcpServers": {
    "signetry": {
      "command": "python",
      "args": ["-m", "signetry_core.mcp_server"],
      "env": { "SIGNETRY_MCP_ROOTS": "${workspaceFolder}" }
    }
  }
}

Then the agent can call signetry_admit, signetry_verify, and signetry_provenance. SIGNETRY_MCP_ROOTS scopes the server to your workspace so it can't be pointed at arbitrary host paths.

2. Project rule (defense in depth)

Drop signetry.mdc into .cursor/rules/ so the agent is told to stay within the contract and to run signetry guard before writing forbidden paths. This is advisory (the model may still err) — the durable guard is running Signetry in CI on the PR via the Signetry Admission GitHub Action.

Note: Cursor has no deterministic pre-write hook like Claude Code, so in Cursor the strong enforcement is the CI check on the PR; the MCP tools + rule give the agent an in-editor way to self-check first.