5m-mcp

A single FiveM resource that exposes a safe MCP (Model Context Protocol) surface, so any MCP-aware agent — Claude Code, Codex, Cursor — can build, debug, and live-test FiveM resources against a running server.

The project is 5m-mcp; the resource you install is named agent_api.

Flagship abilities

  • One agent, one URL. Talks real MCP over HTTP (or a stdio shim). No external sidecar — drop the folder, ensure, paste the printed config into your client.
  • Full dev loop on a live server. Read / write / surgical-edit / multi-edit files in a sandbox, scaffold resources, ensure/restart with console capture, then wait_for_console + scan_errors to catch failures and fix them — without leaving chat.
  • Navigate any resource. Recursive file tree, glob, and substring/regex code search (build output skipped by default).
  • Live in-game testing. Opt-in players, invoke any client/server native by name, and drive the NUI over CDP — screenshot, click, fill, and eval, reaching even cross-origin resource iframes.
  • Framework-aware. Bundled ESX / ox_lib / oxmysql bridges (reflective method calls, DB schema introspection, gated migrations) plus calling any resource's exports.
  • Teach it your way. A web dashboard where you set Preferences (mirror your structure / coding / UI conventions) and upload Skills that auto-inject into the tool calls you choose.
  • Watch & steer the work. A dashboard Sessions board shows the agent's current task + todo list per resource, with a file tree — click any file to send the agent a targeted request, which it picks up on its next tool call.
  • Guided workflows. MCP prompts for the hard parts: scaffold (grill Q&A), debug-resource, add-db-table, and a security-review cheat-prevention audit.
  • Monitor + audit. Realtime, colorized console and a full audit trail of every agent action — live in the dashboard.
  • Safe by default. Auto-generated token, read-only by default (opt into writes), path sandbox, ACE-aware lifecycle, native/SQL blocklists, per-resource locks, and a JSONL audit log of every call.

Install

  1. Drop the folder at resources/[agent]/agent_api/ (or npm run generate:resource).
  2. In the server console: refresh → ensure agent_api.
  3. Copy the MCP config the resource prints on first start into your client.

Full setup, the complete tool catalog, convars, dashboard, and security model live in the docs.