Let any MCP agent
build, run, and debug
FiveM resources live.
A single FiveM resource that exposes a safe, sandboxed Model Context Protocol surface. Claude Code, Codex, Cursor — whatever you use, point it at one URL and ship code straight from chat.
Three lines, then you're done
# FiveM server console
refresh
ensure agent_api
# console prints a copy-paste-ready Claude/Cursor/Codex MCP config blockEverything the agent needs, nothing it doesn't.
Designed for one job: shipping FiveM code with an LLM at the keyboard. Every surface is tracked, gated, or sandboxed.
One drop-in FiveM resource
No external sidecar. Drop the folder into resources/[agent]/, ensure agent_api, paste the printed MCP config into your client.
36+ tools out of the box
Filesystem (sandboxed), resource lifecycle, console capture, live-player probes, dynamic client + server natives, allowlisted shell.
Framework bridges
ESX, ox_lib, oxmysql plugins auto-detect on boot. Reflective dispatchers expose every method behind a readonly verb gate.
Speaks real MCP
HTTP transport for Claude Code, Cursor, and any spec-compliant client. Bundled stdio shim for clients that need stdin/stdout.
Live UI screenshots
Attach to FiveM CEF over CDP, isolate one resource iframe, capture a clean PNG, hand it back to the agent for review.
Built for safety on a live server
Token bootstrap, per-resource locks, readonly mode, blocklists for dangerous natives, ACE-aware, audit log of every action.
Read the docs. Wire your agent. Ship.
Pick your stack and follow the integration guide — every supported client has its own page with a copy-paste config block.