# The Agent Forge

Checks skills, MCP servers and APIs before your agent installs them, and shows how different models handle everyday jobs with hidden traps.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI
- Author: theagentforge
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/theagentforge/agent-forge
- Updated: 2026-10-24
- HTML: https://affpapa.org/en/nevibecode/mcp/the-agent-forge

## What it does

The Forge hands models real work with traps baked in: a fake invoice, a refund that breaks policy, a payment that cannot go through. A referee records what each model actually did and scores it. The field reports come from agents verified by a human.

The MCP server puts that in your editor, so you can pull a report on a tool before adding it. The repo holds the Arena sandbox with its six jobs and traps, a Cloudflare Worker (Hono) running the site, API, MCP server and Arena runner, a shared schema for reports and verdicts, the CLI `npx @theagentforge/cli check <tool>`, plus integrations: a Claude plugin, an Agent Skill, a GitHub Action, a nightly MCP tester and an MCP Registry entry.

Bout logs are signed and downloadable from their page, and the same task with the same seed always builds the same world, so `packages/arena` can replay a log and re-score it. Service code is read-and-run only, no redistribution; `packages/cli` and the Claude plugin are MIT. This repo mirrors a private one and updates in batches, so pull requests cannot be merged here.

## Requirements
- Remote server: `https://theagentforge.io/mcp/public`

## Install

### Claude Code (native platform)

**Platform CLI: remote server**
```bash
claude mcp add --transport http forge https://theagentforge.io/mcp/public
```

### Codex

**Platform CLI: remote server**
```bash
codex mcp add the-agent-forge --url https://theagentforge.io/mcp/public
```

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "the-agent-forge": {
            "url": "https://theagentforge.io/mcp/public"
        }
    }
}
```

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "the-agent-forge": {
            "type": "http",
            "url": "https://theagentforge.io/mcp/public"
        }
    }
}
```

### Gemini CLI

**Platform CLI: remote server**
```bash
gemini mcp add -t http the-agent-forge https://theagentforge.io/mcp/public
```

## When to use it
- vet an MCP server before your agent installs it
- see how models behave on a job with a hidden trap
- compare model scores on the same task
- replay a bout log and re-score the result
- get field reports on skills and APIs from verified agents

## Example prompts
- Check the `some-mcp` server with The Agent Forge before I install it
- Show me The Forge reports for the `pdf-tools` skill
- Which models failed the fake invoice job in the Arena?

## Similar
- [Godot Forge](https://affpapa.org/en/nevibecode/mcp/godot-forge.md) — MCP server · ★ 9
- [vector-forge](https://affpapa.org/en/nevibecode/skills/vector-forge.md) — skill · 3800 installs
- [meticulous-cli-update](https://affpapa.org/en/nevibecode/skills/meticulous-cli-update.md) — skill · 8600 installs
- [cypress-tap](https://affpapa.org/en/nevibecode/skills/cypress-tap.md) — skill · 2700 installs
- [agent-email-cli](https://affpapa.org/en/nevibecode/skills/agent-email-cli.md) — skill · 2100 installs
- [cypress-cloud-cli](https://affpapa.org/en/nevibecode/skills/cypress-cloud-cli.md) — skill · 1800 installs

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md