mcpField
O
orbit

orbit

Local
github.com/constellation-worksAI & Agents17Updated today
Repository

About orbit

Orbit is a local-first runtime for coding agents that manages tasks and controlled delivery. It combines isolated worktrees, locks for parallel runs, approval gates, and audit records without relying on cloud services or receiving provider API keys.

Key features

  1. 01Durable task queue and audit records
  2. 02Isolated worktrees and file locks
  3. 03Gated plan, execute, and review flow
  4. 04Delivery workflows and pull request support
  5. 05Parallel agent orchestration
  6. 06Dashboard, plugins, and agent skills

Use cases

  1. 01Turn requests into tasks with acceptance criteria
  2. 02Run multiple coding tasks in parallel
  3. 03Review agent changes before merging
  4. 04Trace task prompts, execution, and review

MCP support

Orbit can connect to MCP and provides MCP tools, each with a CLI counterpart. Its documentation also describes federated MCP across multiple machines.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
  "mcpServers": {
    "constellation-works-orbit": {
      "command": "npx",
      "args": [
        "-y",
        "@orbit-tools/cli"
      ]
    }
  }
}
Run commandnpx -y @orbit-tools/cli

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npm@orbit-tools/cli0.19.0

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Orbit

Agents write. Orbit delivers.

Release npm License: MIT Docs

Animated tour of the real Orbit dashboard while Orbit's agents work on Orbit itself: a proposed task waiting for approval, 16 tasks running at once across two hosts with 100 files locked so overlapping work waits its turn, an auto-drain with 12 running and 19 admitted that pauses new starts while the host runs hot, and auto-concurrency forecasting CPU and memory for each admission; a Codex pull-request pipeline run from isolated worktree through implement, commit, validate, review gate and push to an open pull request, 21 steps in 24 minutes 15 seconds; routines on a host clock and 22 auto-task definitions that keep the queue fed; analytics showing 3,277 tasks done in 30 days, 167% more than the 30 days before, and each model's implementation times; and the codebase view ranking the paths agents struggle with across 6,002 tasks.

Orbit is a local-first runtime for coding agents. Keep using your authenticated agent CLI; Orbit adds a durable task queue, sandboxed worktrees, file locks for parallel runs, a gated delivery pipeline, and an audit log.

Why: fast agents make planning, review, and traceability easy to lose. Orbit keeps the prompt, plan, and review behind every change, with a task ID on every workflow commit.

  • Single binary, no cloud. State lives in ~/.orbit and .orbit/; Orbit sends no telemetry.
  • Bring your own agents. Use the provider CLIs you already have authenticated, without giving Orbit API keys.
  • MIT licensed. No paid tier or hosted offering.

How it works

You:    The fsProfile lookup is undocumented. Get that fixed.
Agent:  Files a proposed task with acceptance criteria. Approve it and ship?
You:    Yes.
Agent:  Queues it, then runs plan → execute → review in a locked, isolated worktree.
        Opens a pull request. Review and merge it, then approve the task to close it.

Nothing starts without approval; PR runs stop for your review unless you authorize --complete. The task record keeps the prompt, plan, execution trace, and review. For larger specs, ask your agent to orchestrate: the bundled orbit-orchestrate skill splits the work, queues it once approved, runs it in parallel, and diagnoses failures (delivery workflows).

Quick start

You need: macOS or Linux (x64 or arm64; Windows uses WSL2), Node 18+, an authenticated agent CLI, and authenticated gh for pull requests.

npm install -g @orbit-tools/cli
orbit init    # machine name, task-ID prefix, agent skills, and Linux sandbox setup

Open your agent in the repo and ask it to “set up Orbit for this repo”. The bundled orbit-setup skill registers it, connects MCP, runs orbit doctor, and asks for choices such as the PR target branch. Start a fresh agent session, then ask: “Add a hello.txt file containing hello; file it as an Orbit task.” Approve and ship the task when prompted, review and merge the PR, then approve the task to close it. Watch with orbit web serve.

Prefer a plugin? Install Orbit for Claude Code, Codex, or Cursor. Every MCP tool has a CLI twin; for the same loop without an agent, see your first task from the terminal.

Set up the repo by hand 
cd <repo> && orbit workspace init --mcp    # add --ship-mode local to skip PRs
orbit doctor
orbit web serve

Review and commit the checkout files listed by workspace init, including MCP client files, before the first ship. Local delivery refuses tracked changes, merge conflicts, and untracked paths overlapping incoming changes.

orbit doctor checks CLI presence for default, system, and complexity-routed crews and warns when no MCP client is registered; it does not check provider sign-in or MCP connectivity. Use orbit doctor providers for all executor definitions. See Linux sandbox readiness and distro coverage.

To…Run
Inspect a task or runorbit task show <ID> · orbit run show <RUN_ID>
Open the dashboardorbit web serve (remote: orbit web connect <host>)
Open the dashboard in a desktop windowBuild the app in desktop/
Pick the default crew (provider and model)orbit config set workflow.default_crew <crew>
Route tasks to crews by complexityorbit config set --global workflow.hard_complexity_crews '["opus"]' (more)
Upgradenpm install -g @orbit-tools/cli@latest (orbit update --check shows what's new)

Features

Crews and complexity routing

Crews name a provider, model, and effort level; complexity pools choose one for a task without an explicit crew. See crew definitions, init's seeded pools, and complexity routing, weighting, and overrides.

MCP and authority

Connect your agent explains operator and agent-only authority; federated MCP connects multiple machines.

Where state lives

Scoping rules explain machine and workspace state. See the runbooks for backups, recovery, and upgrades.

Learn more

Contributing

Pull requests are welcome, from typo fixes to new executors. Small fixes can go straight to a PR; bigger changes start with an issue. See CONTRIBUTING.md.

License

MIT

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.