# Bouncer

> A governed gateway for AI agents, APIs and MCP servers, plus a free Security Check that vets npm packages and remote MCP servers before you install them. Free to use, no account needed for a first check.

Bouncer sits between your agents and the APIs or MCP servers they call. Register an agent or paste an OpenAPI spec and you get one proxy URL with allow/deny policies, rate limits, spend caps, a full call log and optional per-call x402 payments. The Security Check is static only: package metadata, tarball contents and MCP tool definitions are read as text and never executed.

## Start here

- [Home](https://bouncer.run): What Bouncer is, who it is for, and the two sides of the product (gateway and Security Check).
- [Security Check](https://bouncer.run/check): Paste an npm install command, a package name like `axios@1.20.0`, or an MCP address like `https://mcp.example.com/mcp` and get a verdict, risk score and plain-language findings.
- [Pricing](https://bouncer.run/pricing): The gateway is free; x402 per-call payments go straight from caller to API owner with no platform cut.
- [API catalog](https://bouncer.run/catalog): Every API published on the gateway, with its description, proxy URL and per-call price.

## Security Check

- [Run a check](https://bouncer.run/check): Detects install scripts and what they actually contain, real obfuscation, credential access, outbound domains, known advisories, and hidden instructions or invisible characters in MCP tool descriptions.
- [npm report format](https://bouncer.run/check/npm/axios/1.20.0): Report URLs follow `https://bouncer.run/check/npm/<name>/<version>`.
- [MCP report format](https://bouncer.run/check/mcp/mcp.deepwiki.com/mcp): Report URLs follow `https://bouncer.run/check/mcp/<host>/<path>`.
- [Public check API](https://bouncer.run/api/public/v1/check?ecosystem=npm&name=axios&version=1.20.0): `GET /api/public/v1/check`, `POST /api/public/v1/scans`, `GET /api/public/v1/scans/:id`, `GET /api/public/v1/config`. Rate limit 10 checks per hour per IP.

Notes: private MCP servers accept a one-time token that is used for that single check and never stored, and token-checked reports stay private. Re-scans with the latest rules require a signed-in account; first checks are open to everyone. A verdict is an informed automated opinion, not a guarantee.

## Gateway

- [Add an agent](https://bouncer.run/auth): Sign up, add an agent, paste an OpenAPI/Swagger URL or set a manual proxy target, then set policies, limits and optional pricing.
- [Proxy endpoint](https://bouncer.run/api/public/gw/): Callers use `https://bouncer.run/api/public/gw/<slug>` with the header `x-gateway-key: <key>` instead of the real backend URL.
- [What is an MCP gateway?](https://bouncer.run/mcp-gateway): The architecture: allow lists, rate limits, spend caps and audit logging in front of MCP servers and REST APIs.
- [Blog: Hello, agent!](https://bouncer.run/blog/hello-agent): Why the gateway exists and what it governs.

## Tools for agents

- [WebMCP documentation](https://bouncer.run/webmcp): Tools exposed on every page under `window.__webmcp.tools`, and the OAuth-protected MCP endpoint at `https://bouncer.run/mcp`.
- [MCP security guide](https://bouncer.run/mcp-security-guide): Main MCP security risks (hidden tool instructions, tool changes after approval, shadowing, over-broad access) and best practices.
- Public tools: `security_check_scan` (`input`, optional `mode` = `npm` | `remote_mcp`, optional `auth_token`), `security_check_get_report` (`name`, optional `version`, optional `ecosystem`), `security_check_recent`, `bouncer_overview`, `bouncer_navigate`, `bouncer_how_to_publish_api`.
- Account tools (after sign-in): `bouncer_list_apis`, `bouncer_get_api`, `bouncer_list_calls`, `bouncer_usage_summary`, `bouncer_create_api`, `bouncer_add_policy`, `bouncer_set_api_active`.

## Legal

- [Privacy Policy](https://bouncer.run/privacy): What is collected, how long it is kept, and your GDPR and CCPA rights.
- [Terms of Service](https://bouncer.run/terms): Acceptable use, what a verdict does and does not promise, liability.
- [Sub-processors](https://bouncer.run/subprocessors): Every third party that may process customer data, and how to request a DPA.

## Optional

- [x402 spec](https://x402.org): The per-call payment protocol Bouncer supports.
- [Sandbox facilitator](https://x402.org/facilitator): Test payment settlement without real money.
- [Agent instructions](https://bouncer.run/AGENTS.md): Short machine-readable guidance for agents operating on this site.

Contact: nader@cserny.com