Agent frameworksWorkflow automation

OpenWork

different-ai/openwork

Share one set of AI skills, MCP servers, and connected accounts across every agent and every teammate.

Repository
Stars
21.7k
Forks
1.6k
Open issues
96
Language
TypeScript
Licence
See repo
Created
2026-01-14
Last push
2026-08-09
Effort to adopt

Some wiring

A few days of integration work — credentials, data mapping, a deploy target.

01

What it actually does

Most teams that adopt AI agents end up with the same mess: one person wires a useful skill into Claude Code, someone else rebuilds it in Cursor, a third does it again in Codex, and none of them can use each other's work. Credentials get pasted into three places. Nothing is governed.

OpenWork is a desktop app and, more importantly, a single MCP server that fixes the sharing problem. You define a capability once — a skill, a plugin, an MCP connection, a Google Workspace or Microsoft 365 integration — and then any compatible agent can reach it. The MCP exposes just two tools: `search_capabilities` to find what is available, and `execute_capability` to run it.

The desktop app is optional. If your team already lives in Claude Code or Cursor, you add the MCP and keep working where you are. For larger organisations there is an admin interface to publish capabilities, manage who can use what, and configure whether a connection is shared or per-user.

It is open source and positions itself as the alternative to Claude Cowork, built on opencode, running on macOS, Windows, and Linux.

02

Who it's for

  • 01

    Teams where three different people have each wired up the same integration in three different agents

  • 02

    Ops leads who need to control which agents can touch the CRM, and with whose credentials

  • 03

    Founders who want one place to see every AI capability the company has built

  • 04

    Agencies that need to reuse the same workflow across multiple client workspaces without copy-paste

OperationsFoundersEngineeringMarketingSaaSAgencyAny industry
03

Where it earns its keep

  • Publishing an approved 'lookup a customer in the CRM' capability that everyone's agent can call
  • Sharing Google Workspace and Microsoft 365 access with agents without handing out raw credentials
  • Standardising a set of skills across a team that uses a mix of Claude Code, Cursor, and Codex
  • Giving a new hire working agent capabilities on day one instead of week three
  • Keeping per-user connections separate so each person's agent acts as them, not as a shared service account
04

Use it, or skip it

Reach for it when

  • More than two people are building agent workflows and the work is not being reused
  • You need an audit trail of which capabilities exist and who can run them
  • Your team is split across different agent clients and you refuse to standardise on one
  • You want to keep credentials out of individual config files

Skip it when

  • You are one person with one agent — the coordination layer solves a problem you do not have yet
  • You need a hosted, vendor-supported product with an SLA rather than something you run yourself
  • Your compliance posture forbids a desktop app holding connection configuration for shared services
  • You only need one integration and a plain MCP server would do the job with less moving parts
05

10 automations

  1. 01Operations

    Company capability catalog

    Publish every approved automation as a capability, so anyone can ask their agent what's available instead of asking in Slack.

  2. 02Sales

    Shared CRM lookup for the whole team

    One governed capability that reads deal state from the CRM, callable from any agent, with per-user credentials so access rules still apply.

  3. 03Operations

    New-hire agent onboarding

    Grant a role-based capability bundle on day one — marketing gets the content and analytics tools, sales gets CRM and outreach.

  4. 04Operations

    Client workspace templates

    For agencies: define a standard capability set per client engagement and clone it, rather than rewiring integrations for every account.

  5. 05Founders

    Weekly capability usage report

    Track which capabilities actually get executed and which were built and forgotten, then retire the dead ones.

  6. 06Engineering

    Credential rotation without chaos

    Rotate a shared integration's credentials in one place and have every teammate's agent keep working without touching a config file.

  7. 07Operations

    Approval-gated capabilities

    Mark destructive capabilities (send email, update a deal, post publicly) as requiring explicit human approval before execution.

  8. 08Marketing

    Cross-tool content workflow

    Draft in one agent, fact-check in another, publish through a shared capability — the same skill definitions apply throughout.

  9. 09Support

    Support macro library

    Turn your best support responses into capabilities so every agent answers consistently, with the current docs as the source.

  10. 10Operations

    Capability changelog to Slack

    Post a message whenever someone publishes or changes a shared capability, so the team knows what their agents can suddenly do.

Want one of these running by Friday?

LimeDock builds these as real workflows inside your stack — deployed to your cloud, wired into your Slack and CRM, with the code in your repo. You pay a build fee and your own API keys, nothing else.

Book a workflow call
07

Source

Repository stats were read from the GitHub API and reflect the last time we refreshed this entry. The editorial breakdown above is LimeDock’s own analysis — we are not affiliated with different-ai.

https://github.com/different-ai/openwork