← Discover MCPs and Agents
u
MCPAI & MLGitHub

useagent

The open-source AI coworker for your team: agents with their own cloud computer, your tools and context, handing back finished work websites, decks, spreadsheets, reports, PRs. Runs Claude Code, Codex, OpenCode on your subscription.

Links

README

From the repo.

useAgent

The open-source AI coworker for your team.
Your agents. Their own computer. Finished work you can use.

Status: alpha License: AGPL-3.0 Runtime: bun Documentation

Watch the demo · Quick Start · Self-hosting · Documentation · Architecture

useAgent gives Claude Code, Codex, OpenCode, and Pi a shared workspace with repositories, a terminal, a browser, and your team's tools and context. Ask for research, a website, a spreadsheet, or a code change. Follow the work in the thread, step in when needed, and open the files it produces.

Watch useAgent in action: an agent session with a live timeline, terminal, and workspace panes

Watch the 63-second product tour →
Real product footage. No sign-in required. The recorded UI may differ from your release.

Alpha software. useAgent is under active development: expect rough edges, and APIs/schemas may change between releases. It already runs real daily workloads, but pin a tag if you need stability.

Features

A computer for every thread

Watch an agent research in a real browser, run commands, and work with your repositories. Open its desktop or terminal and take control when you need to. Daytona and CubeSandbox provide isolated Linux workstations with screen recording.

Computer use and sandboxes →

An agent reads browser sources while its tool activity and research stream into the thread

Teach it how your team works

Import skills from GitHub, pin a playbook to a task, and reuse the procedures that work. Skills are versioned, so a run records exactly what it used.

Skills and playbooks →

Browse reusable skills and the team's knowledge from the shared workspace

Context that carries forward

Keep knowledge, wiki pages, and optional team memory beside the work. Inspect recalled facts, correct them, and keep personal and organization memory separate.

Knowledge and memory →

The memory hub with separate organization and personal views, shown in dark and light themes

Also in the workspace:

  • Files you can use - documents, spreadsheets, presentations, PDFs, images, and videos. Supported workpieces have revisioned edits and native exports.
  • Work from Slack - mention an agent, send attachments, and receive artifacts in the thread.
  • Recurring work - schedule tasks, inspect their history, or run one immediately.
  • Approval controls - gated tools pause for a decision and resume with a one-shot, argument-bound capability.
  • Durable sessions - Postgres stores the event timeline; recovery re-probes live sessions after a restart.

Supported agents

Claude Code   Codex   OpenCode   Pi

One session UI and event contract across engines. Connect a supported provider account or API key; available models, login methods, and tools depend on the adapter and your self-hosted configuration. See the engine guide.

Integrations

Work arrives from anywhere and tools stay behind the gateway:

SurfaceWhat's connected
Channels inWeb app, Slack, REST API, schedules - every channel enters through the same run door
NativeSlack (mentions, threads, delivery), GitHub (App auth, clones, PRs)
Via connectorsGmail, Linear, Notion, HubSpot - OAuth handled by the broker, tokens sealed server-side
Workspace surfacesKnowledge base, team memory, skills and playbooks, scheduled automations

Quick Start

Clone the public repository:

git clone https://github.com/useagenthq/useagent.git
cd useagent

Requires bun and Postgres 16+ with the pgvector extension (stock Postgres images do not include it). No Postgres handy? One container does it:

docker run -d --name useagent-pg -p 127.0.0.1:5432:5432 \
  -e POSTGRES_HOST_AUTH_METHOD=trust pgvector/pgvector:pg16
export DATABASE_URL=postgres://postgres@localhost:5432/postgres
for workspace in \
  packages/agent-harness packages/artifact-workspace \
  packages/agent-client packages/artifact-formats packages/sandbox-contract \
  packages/conformance packages/cli \
  backend frontend; do
  (cd "$workspace" && bun install --frozen-lockfile)
done

bun run dev:backend    # API + orchestration on :3201
bun run dev:frontend   # UI on :3400 (proxies /api/* to the backend)

bun run typecheck covers every package.

The database example is for local development. See the setup guide for provider credentials and sandbox configuration before running a real agent task.

Self-hosting

useAgent runs on any Linux host - AWS, Google Cloud, Azure, or bare metal. See infra/self-host/ for the full guide, including the one-command reference host (Terraform) and the provider-agnostic deploy-app.sh:

SERVER_IP=<host-ip> PG_PASSWORD=... OPENROUTER_API_KEY=... \
  infra/self-host/deploy-app.sh /path/to/this/repo

Sandboxes are pluggable: Daytona (managed service - pairs with a host on any cloud, easiest start) or CubeSandbox (self-hosted runtime on your own hardware - full data locality). Production deploy lanes are documented in infra/self-host/; the provisioning is provider-agnostic and addresses the host over SSH.

Architecture

useAgent architecture: entry channels feed a self-hosted control plane (Run API, Postgres event log, engine adapters, session UI); adapters spawn an isolated cloud sandbox per thread; every integration crosses the trusted gateway; finished work comes back as editable artifacts

Three properties do the heavy lifting:

  1. The engine is a plug. Claude Code, Codex, OpenCode, and Pi all speak one canonical event contract through the engine adapters - swap engines and your threads, artifacts, and memory stay.
  2. Every run is an event log. Postgres is the source of truth: runs survive backend restarts, replay exactly, and stay inspectable after the fact.
  3. Credentials never enter the sandbox. The agent's computer is isolated; every integration call crosses the trusted gateway as a typed tool, and the keys live only on your control plane.
PathWhat it owns
frontend/Product UI: chat, sessions, skills, playbooks, wiki, artifacts, automations, settings
backend/Control plane: auth, runs, sandboxes, engines, knowledge, memory, artifacts, connectors
packages/Shared contracts: thread events, canonical engine events, workpieces, renderers
docs-site/Documentation site: concepts, architecture, API, operations
infra/self-host/Self-hosting on any provider, with a reference Terraform host
memory/Optional team-memory service

Deeper reading: the documentation site and the interactive request-flow diagram.

The additive immutable-container release lane is documented in docs/operations/immutable-releases.md.

Community and contributing

License

useAgent is free and open-source software under the GNU AGPL v3.0 (AGPL-3.0-only). You may use, modify, and self-host useAgent under the AGPL.

If you want to embed useAgent into proprietary software, distribute it without AGPL obligations, build an OEM or white-label product, or obtain different terms, a commercial license is available.

Contributions are accepted under the CLA. The useAgent name and logo are covered by the trademark policy, not the code license. Third-party components are listed in NOTICE; vendored and ported files carry per-file attribution headers.

Collected info

  • 346 stars
  • 62 forks
  • Language: TypeScript
  • Source updated: 9/24/2026

Config for your environment

Replace {MCP_ENDPOINT_URL} with this MCP’s endpoint URL (from its repo or docs above). No API key — you connect directly.

Tool

OS

Config file: ~/.cursor/mcp.json

{
  "mcpServers": {
    "mcp-server": {
      "url": "{MCP_ENDPOINT_URL}"
    }
  }
}

Paste into mcpServers in the config file. Restart Cursor after saving.

If this MCP is also published on mcpchannel.ai, you can subscribe from Browse and use the gateway config there instead.