SkillFoxx

FleetQ MCP

An agent orchestration platform with a 670-tool MCP server: Claude Code, Cursor and Codex drive DAG workflows and agent crews.

  • 81 ★ on the repository
  • Last commit
  • AGPL-3.0
MCP server

Broad access

Why this level

  • The platform and its MCP server can run bash, connect over SSH, make git pushes and deploys, so an outside agent gets broad access.
All reasons and checks

escapeboy/agent-fleet-oAuthor:escapeboy

Install

Manual install

git clone https://github.com/escapeboy/agent-fleet-o.git

Then `make install` starts Docker and the setup wizard. Connect the MCP client at the `/mcp` address of your install.

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • The platform and its MCP server can run bash, connect over SSH, make git pushes and deploys, so an outside agent gets broad access.

This is third-party code. Review the repository files before installing.

What it does

FleetQ is a self-hosted agent control panel on Laravel: visual DAG workflows with eight node types, agent crews with coordinator, worker and reviewer roles, human approvals, budgets and an audit trail. Every platform capability is open over MCP: the `AgentFleetServer` server gives 670+ tools across 60+ domains over stdio, HTTP/SSE and OAuth2 with PKCE, plus a REST API with 175+ endpoints. An outside agent connects to the platform, obtains credentials through dynamic client registration and manages workflows, experiments and tools. The repository ships the `fleetq-connect`, `fleetq-manage-workflow` and `fleetq-run-experiment` skills. Actions with a real effect, such as integration writes and git pushes, go through a risk policy with tiers: auto, ask or reject.

Who it is for. Teams building multi-step agent systems who want to drive them from Claude Desktop, Cursor or Codex rather than writing an orchestrator.

Good fit when

  • A multi-agent workflow is built on a graph with conditions, parallel branches and manual approval.
  • A Claude Code agent starts an experiment and fetches the result through `fleetq-run-experiment`.
  • A budget is set on a project and the platform pauses work on overspend.
  • Agent writes to integrations are approved by a human according to risk tier.

Not a fit when

  • You need a single tool for a single script, while the platform is heavy and needs PHP, PostgreSQL and Redis.
  • The service cannot be self-hosted and the cloud version does not fit.
  • The AGPLv3 license is unacceptable to you.

Example request

Connect to my FleetQ, list the workflows and start an experiment comparing two prompt variants, waiting for approval.

MCP

Transport
stdio, http
Authentication
OAuth

Limitations

The platform is installed on your own server: `make install` in Docker, or by hand with PHP 8.4, PostgreSQL 17, Redis 7 and Node.js 20. The `APP_AUTH_BYPASS=true` mode turns off the password and suits only a local machine with `APP_ENV=local`. The AGPLv3 license requires publishing changes when the service is offered over a network. The 670-tool set is broad, so connect an outside agent with minimal rights and keep the approval policy on: the platform can run bash, connect over SSH, make git pushes and start deploys. A managed FleetQ Cloud version exists.

How to disable. Disconnect the MCP server in the client configuration and stop the platform with Docker Compose commands. Revoke tokens and OAuth clients in the FleetQ team settings.

FAQ

How do I connect Claude Code or Cursor?

Use the HTTP/SSE transport at the `/mcp` address of your install or stdio. The `fleetq-connect` skill describes dynamic client registration and obtaining a token.

How do I get a token without manual setup?

Through RFC 7591: fetch metadata at `/.well-known/oauth-authorization-server`, register a client at `/oauth/register` and run the authorization code flow with PKCE. Or create a Sanctum token in team settings.

What is needed to install it?

Docker and `make install`, or PHP 8.4, PostgreSQL 17, Redis 7, Node.js 20 and Composer. The setup wizard creates the database, an admin account and an LLM provider.

Can dangerous actions be limited?

Yes. The action policy sorts assistant calls, integration writes and git pushes by risk tier: run, ask or reject. Approved actions run automatically.

Which models are supported?

Anthropic, OpenAI, Google, plus Ollama, LM Studio, vLLM, Codex and Claude Code with your own key. There is no single-vendor lock-in.

Discussion

No questions or reviews yet.

MCP servers

UseAgent’s MCP server sends tasks to remote agents, collects their results and records the runs in an organization’s ledger.

MCP serverBroad access382

A CLI that runs independent coding-agent sessions in tmux or herdr and links them through durable messages on a shared broker.

CLIBroad access56

A CLI that plugs third-party models into Claude Code orchestration: workflows, agent teams and subagents on DeepSeek, GLM, Kimi, Qwen or a Codex subscription

CLIBroad access190

A self-hosted runtime for coding agents: each crew in its own Linux container, wrapped in a control plane with roles and audit

CLIBroad access5

A layer over coding agents that assembles a temporary specialist team per task inside Claude Code, Codex, Cursor and Gemini

PluginExtended access1.5K
Official

Preloop is an open source agent control plane: an MCP firewall, a budgeted model gateway and event-driven agent runs.

MCP serverBroad access96

README in short

The README sets out the purpose of the platform, key concepts, features, use cases, a comparison with alternatives, Docker and manual installation, authentication, configuration, SSH access, architecture and the MCP server.

README badge

Report a problem