Git-native policy layer for AI agents: check_action verdicts against rules approved via PR.
Copy the AI prompt to install this server into Claude Code, Cursor, or another agent β or use 1-click editor setup below.
π‘ Paste the JSON block into your client's configuration file under mcpServers, then restart the application.
Run the setup below once and every agent your team runs has a git-native rulebook it has to check over MCP before it acts, approved the same way your code already is: a merged pull request.
check_action over MCP before anything risky (a refund, a delete, a message to a customer) and get back an allow/block/escalate verdict.Prefer not to run any of this yourself? The hosted version at gntai.dev does the same thing without you standing up a Postgres instance.
To see a real check_action response before installing Node or npm, clone
the repository and run the Docker-only demo:
It builds the full local stack in an isolated gnt-demo Compose project,
seeds an approved refund rule, calls check_action, and prints a curl command
you can run again. Without an Anthropic key the real fail-closed path returns
needs_human; pass GNT_DEMO_ANTHROPIC_API_KEY=sk-ant-... ./demo.sh to opt in
to the model's grounded policy verdict. The demo uses deterministic local embeddings and
throwaway localhost-only secrets, so it does not need a ZeroEntropy key.

Nothing for
gnt prebrainto scan yet? Rungnt initfirst β it scaffoldsrules/locally with a couple of example rule files so there's something real to look at and edit, and points you atgnt prebrain --starter-packsfor a curated pack to start from instead.
That merge lands a rule file in your connected repo, shaped like this:
Each file is plain markdown with YAML frontmatter:
There's no captured transcript to show yet (see the gap noted at the bottom of this README).
Here's the actual response shape a check_action call returns, straight from the tool's
contract:
verdict is one of allowed, blocked, or needs_human. needs_human is the fail-closed
default: no approved rule covers the action, retrieval failed, or the check couldn't complete.
It never guesses.
One MCP endpoint, five tools:
| Tool | What it does |
|---|---|
check_action | Checks a described action against your approved rules before an agent takes it. Returns allowed, blocked, or needs_human with cited rules and a one-line reason. |
search_rules | Semantic search over your org's approved rules, optionally filtered by tag. An empty list means no approved rule covers the query. |
get_rule | Fetches one approved rule by id, with its provenance (who approved it, when, what it was cited from). |
list_skill_packs | Lists every compiled skill pack version for your org, newest first. |
get_skill_pack | Fetches a compiled skill pack's manifest and file list by id. |
| Requirement | Check | Get it |
|---|---|---|
| Node >=22.13 | node --version | nodejs.org |
| Method | Command |
|---|---|
| curl | curl -fsSL gntai.dev/install.sh | sh |
| npm | npm install -g @gnt-ai/cli |
gnt needs Node >=22.13. If the CLI fails to start with a version error, update Node first and confirm with
node --version.
| Variable | Default | What it controls |
|---|---|---|
GNT_API_URL | https://api.gntai.dev | API endpoint the CLI and MCP calls hit |
GNT_WEB_URL | https://gntai.dev | Web app used for gnt login's browser step |
GNT_CONFIG_DIR | ~/.gnt | Where credentials.json and local config live |
gnt prebrain's default extraction mode is cloud, not on-device: your source text goes straight to Anthropic's API (or Vercel AI Gateway with zero-data-retention, if you configure it), never to gnt's own servers. Fully on-device extraction needs --mode local against a local Ollama daemon.apps/api only sends error data to Sentry if you set SENTRY_DSN yourself. Leave it unset and nothing goes out.gnt prebrain (batch-extracted from real sources) or gnt review (hand-proposed).rules/<rule-id>.md files with the frontmatter shown above and a plain markdown body.gnt rules lint checks a rule file's frontmatter locally, and gnt-ai/gnt/.github/actions/lint-rules runs the same check as a CI step on your rules repo's own PRs, so a bad frontmatter fails the PR instead of the review.Self-hosting:
gnt login's browser step has nowhere to land.gnt loginopens a browser to a/cli-loginpage and polls the API for the resulting key β that page is served by the hosted product's web app, which isn't part of this repo. There's no CLI-only login flow (device code or otherwise) today, and nogntcommand to set a key manually. Self-hosting this stack currently means building your own thin frontend for that one route (it just needs to complete the sign-in flow and hand the CLI a key). This is a real, open gap in the self-host path, not a config issue β closing it properly means adding a CLI-only login flow.
ValueError: refusing to start: these settings still have their .env.example placeholder value...Achange-me-...string is still sitting inapps/api/.env. The error names every offending field; generate a real value for each and retry.
storefails to start withGNT_STORE_INTERNAL_API_SECRET is not set.apps/store/.envwasn't filled in, or wasn't picked up. Confirm the file exists at that exact path, not still named.env.example.
Every store-to-api call gets rejected with 401 or 403, even though both services are up.
STORE_INTERNAL_API_SECRET/APPROVAL_SIGNING_SECRETinapps/api/.envdon't byte-for-byte matchGNT_STORE_INTERNAL_API_SECRET/GNT_APPROVAL_SIGNING_SECRETinapps/store/.env. This fails closed by design. Regenerate both pairs so the two files agree.
A rule fails to save with an embedding or rerank error.
apps/store/.envis missingZEROENTROPY_API_KEY, or it's still empty. Get a real one from zeroentropy.dev.
No reviews yet β be the first to share how this listing worked for you.
Showcase your server listing on GitHub or your project documentation. Embed this dynamic SVG badge to highlight official listing status and live engagement.
[](https://allmcps.com/mcp/gnt)<a href="https://allmcps.com/mcp/gnt"><img src="https://allmcps.com/api/badge/gnt?style=directory" alt="Gnt on AllMCPs" /></a>