apra-fleet

Overview

apra-fleet is an open-source MCP server and CLI for running fleets of AI agents across your own machines, providers, and workflows. Its pitch is that one AI agent is a demo, while many agents across laptops, GPU boxes, VMs, CI machines, and different LLM providers become an operations problem.

The tool lets an MCP-capable orchestrator control a fleet of registered members. A member can be a local or remote machine running a provider CLI or local model. apra-fleet can then dispatch commands, prompts, files, credentials, and workflow steps across that fleet.

For developers already using tools such as Claude Code, Codex, Gemini, Copilot, OpenCode, or Antigravity, apra-fleet is not another single-agent chat interface. It is infrastructure for scheduling, isolation, credentials, observability, and durable multi-agent work. That makes it a strong fit for the AI agent category and for teams experimenting with larger coding-agent systems.

Key Features

• MCP-compatible fleet control - apra-fleet exposes an MCP server so compatible agents can register members, inspect topology, dispatch work, and coordinate workflows.
• Cross-machine execution - Register Windows, macOS, and Linux machines locally or over SSH, then route tasks to the right device or provider.
• Multi-provider routing - The project supports provider strategies for Claude, Codex, Gemini, Copilot, Antigravity, OpenCode, and OpenAI-compatible local endpoints.
• Durable workflows - Workflows are resumable programs with execution journals, member reservations, stall detection, watchdogs, and recovery behavior for longer unattended runs.
• Out-of-band credentials - Secrets are entered outside the model context, encrypted at rest, scoped to members, and resolved server-side during execution.
• Software engineering workflow - The flagship fleet-sprint workflow plans, develops, reviews, tests, and harvests software work across multiple agents.

How to Get Started

1. Install Node.js 22+ - apra-fleet requires a modern Node runtime.
2. Install the CLI - Run npm install -g @apralabs/apra-fleet.
3. Start the fleet console - Use the CLI to launch the control plane and inspect fleet state.
4. Register members - Add local or SSH-accessible machines as members, then pair them with provider CLIs or local endpoints.
5. Configure credentials - Enter secrets out-of-band so models do not see provider keys.
6. Run a workflow - Start with a documented workflow such as fleet-sprint, then author custom workflows as your operational pattern becomes clear.

Community & Ecosystem

apra-fleet is published under the Apache-2.0 license and is hosted on GitHub. The README lists CI, platform support for macOS, Linux, and Windows, MCP compatibility, a documentation site, and a DeepWiki codebase assistant.

The project is also dogfooded: its README states that an autonomous apra-fleet workflow plans, codes, reviews, tests, and ships the apra-fleet codebase itself. Treat that as an early-stage but useful signal that the maintainers are testing the workflow against real repository maintenance rather than only a demo script.

Pricing & Plans

apra-fleet is free and open source. The official site says there is no paid tier and that users bring their own LLM keys.

Your actual operating cost depends on the machines you run, provider CLIs you connect, model tiers, and how many prompts your workflows invoke.

Best For

• Developers who want to orchestrate Claude, Codex, Gemini, Copilot, OpenCode, and local models from one control plane.
• Engineering teams experimenting with multi-agent software development workflows.
• AI infrastructure builders who need durable workflows, scheduling, credentials, and observability.
• Teams comparing AI code generator workflows and wanting to move beyond single-agent runs.
• Organizations that prefer open-source agent infrastructure over a hosted-only orchestration product.

FAQ