Metadata-Version: 2.4
Name: perforce-agentic-gateway
Version: 2026.3
Summary: Local MCP gateway that proxies AI clients to many MCP servers with a token-efficient code mode
Requires-Python: >=3.8
License-Expression: LicenseRef-Proprietary
Keywords: mcp,model-context-protocol,perforce,ai,gateway,proxy,agentic
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: Environment :: Console
Classifier: Topic :: Software Development
Classifier: Topic :: Software Development :: Version Control
Classifier: Programming Language :: Go
Classifier: Operating System :: POSIX :: Linux
Classifier: Operating System :: MacOS
Classifier: Operating System :: Microsoft :: Windows
Description-Content-Type: text/markdown
Project-URL: Homepage, https://github.com/perforce/perforce-agentic-gateway
Project-URL: Documentation, https://github.com/perforce/perforce-agentic-gateway/blob/main/using-pag.md
Project-URL: Changelog, https://github.com/perforce/perforce-agentic-gateway/releases
Project-URL: Issues, https://github.com/perforce/perforce-agentic-gateway/issues

# Perforce Agentic Gateway (PAG)

PAG is a local [MCP](https://modelcontextprotocol.io) gateway with a
token-efficient code mode. You configure your AI client once to connect to
PAG, and manage every other MCP server through PAG: it runs the server
processes, restarts them when they crash, and keeps their credentials in your
OS keychain.

Perforce products ship as first-party servers, and PAG proxies any MCP
server — add third-party servers from a registry or by manual configuration.

## Install

PAG runs via [uv](https://docs.astral.sh/uv/); there is no separate install
step. Point your AI client at it — for example, Claude Code:

```sh
claude mcp add pag -- uvx perforce-agentic-gateway@latest
```

Any client that uses `.mcp.json` (Cursor, and others):

```json
{
  "mcpServers": {
    "pag": {
      "command": "uvx",
      "args": ["perforce-agentic-gateway@latest"]
    }
  }
}
```

Browse the catalog and manage servers in the dashboard with
`uvx perforce-agentic-gateway@latest view`.

## Why it stays fast

Rather than forwarding every downstream tool to your client (20 servers can
mean a thousand tools), PAG exposes a small, fixed `pag__*` toolset and keeps
the full catalog inside the gateway. Your assistant searches it with
`pag__explore` and runs multi-step workflows — loops, fan-out across servers —
through `pag__execute` in one round trip instead of many.

## Documentation

- [Connecting AI clients](https://github.com/perforce/perforce-agentic-gateway/blob/main/connecting-clients.md)
- [Using PAG](https://github.com/perforce/perforce-agentic-gateway/blob/main/using-pag.md)
- [Configuration](https://github.com/perforce/perforce-agentic-gateway/blob/main/configuration.md)
- [Secrets](https://github.com/perforce/perforce-agentic-gateway/blob/main/secrets.md)
- [Troubleshooting](https://github.com/perforce/perforce-agentic-gateway/blob/main/troubleshooting.md)

## License

Use of PAG is governed by the
[Perforce Agentic Gateway Terms of Use](https://github.com/perforce/perforce-agentic-gateway/blob/main/LICENSE.md).
