# Bench47

> Bench47 is an AI software factory built on a development team's context. It keeps the team's decisions, conventions, skills and history in one hosted Markdown workspace that every coding agent reads and writes over MCP: Claude Code, Codex, Cursor, Antigravity CLI, Pi and any other MCP client. It starts agent jobs on the team's own machines when an alert, a ticket, a mention in a thread, an email or a schedule calls for one, and the mb CLI keeps team skills current on every machine.

Bench47 puts your agents to work on tickets, alerts and scheduled tasks. Every run starts from your team's decisions, skills and history, and adds what it learns.

Every plan includes the whole software factory: the team's context, shared skills and Dispatch. Agents run on your own machines with your own subscription or API key; Bench47 doesn't charge for agent usage.

- Team context: Keeps your team's decisions, conventions and history where every agent can read them. Typed records: Keep decisions, incidents or anything else your team tracks as structured records, so agents can find the right ones quickly. Version history: Bench47 keeps earlier versions, so you can see what changed, go back to an older one, or lock something so agents can't edit it. Export: Download a workspace, a folder or a single page as a zip whenever you like. Documents come out as plain Markdown files, so nothing is locked in.
- Long-running work: Pick up long-running work where you left off. Due dates and reminders: Put a date on work you've paused, with a note on how to pick it up. Bench47 reminds you when it's due and hands you the note to give your agent.
- Agent jobs: Start an agent when something happens in your other tools, or on a schedule. Your own prompts: You decide what the agent is asked to do for each kind of event, and you can check the prompt before it runs. Mentions: Mention Bench47 on a Linear issue, a GitHub issue or pull request, or in a Slack thread. An agent picks it up and answers in the thread, as Bench47. Schedules: Run an agent every weekday morning, every hour or once a month, in your time zone. Its prompt lives in a shared doc your team can refine between runs. Runs on your machines: Agents run on the machines you choose, with only the access you give them, and you don't have to open your network to Bench47.

## Connect

- MCP endpoint: https://api.bench47.com/mcp (streamable HTTP; OAuth 2.1 with PKCE and dynamic client registration).
- [Agent documentation index](https://help.bench47.com/agents/index.md): Every help page as plain Markdown, grouped by section.

## Set up with a coding agent

Set me up with Bench47 (https://bench47.com), the AI software factory for my team and our coding agents.

1. Install the Bench47 agent skill. Pick whichever method suits this tool:
   - Recommended, via the skills CLI (https://github.com/AltaCoda/skills), which also lets me update it later with `npx skills update bench47`:
     npx skills add AltaCoda/skills/bench47
   - Or download the single SKILL.md into your skills directory so it ends up at <skills dir>/bench47/SKILL.md, e.g.:
     - Claude Code: mkdir -p ~/.claude/skills/bench47 && curl -fsSL https://raw.githubusercontent.com/AltaCoda/skills/main/bench47/SKILL.md -o ~/.claude/skills/bench47/SKILL.md
     - Codex: mkdir -p ~/.codex/skills/bench47 && curl -fsSL https://raw.githubusercontent.com/AltaCoda/skills/main/bench47/SKILL.md -o ~/.codex/skills/bench47/SKILL.md
     - Antigravity CLI: mkdir -p ~/.gemini/antigravity-cli/skills/bench47 && curl -fsSL https://raw.githubusercontent.com/AltaCoda/skills/main/bench47/SKILL.md -o ~/.gemini/antigravity-cli/skills/bench47/SKILL.md

2. Connect the Bench47 MCP server at https://api.bench47.com/mcp (OAuth sign-in in the browser, no API key). Ask me before adding it to your config, then tell me to finish the sign-in and approve the connection.

3. Once connected, list my workspaces and tell me what to put in Bench47 first for the project I'm working on.

Docs: https://help.bench47.com/humans/mcp/installing

## Pages

- [Home](https://bench47.com/): Turn your coding agents into a software factory.
- [Pricing](https://bench47.com/pricing/): Free: $0 a month (one person); Pro: $20 a month or $192 a year (includes 5 seats); Team: $40 a month or $384 a year (includes 10 seats, then $5 each); Enterprise: quoted (over 100 people). Annual billing is 20% less. Prices are in USD and exclude tax.
- [Compare](https://bench47.com/compare/): Bench47 next to Tessl, Factory and 8090, building your own software factory, and coding agents without one.
- [Blog](https://bench47.com/blog/): Guides, engineering notes and product updates from the team building Bench47.
- [About](https://bench47.com/about/): Why Bench47 exists and who builds it.
- [Security](https://bench47.com/security/): Isolated workspaces, agents you approve and can revoke, history for every change, encryption in transit and at rest, and no training on your content.
- [Contact](https://bench47.com/contact/): Deployment, security review, plan and product questions.
- [Book a demo](https://bench47.com/demo/): Book a 30-minute call with the Bench47 team through Cal.com.
- [Legal](https://bench47.com/legal/): Terms, Privacy Policy, DPA and Subprocessors.

## On every plan

- Skill sync: Publish your team's skills once, and every teammate's coding agent stays on the latest version.
- Reminders: Set a due date on any page. When it comes round, Bench47 reminds you or hands the work to an agent.
- Dispatch: Events from tools like GitHub, Sentry or Linear, a mention in a Linear, GitHub or Slack thread, an email or a schedule you set start a coding agent with the context it needs.
- Schedules: Run an agent every weekday morning, every hour or once a month, in your time zone. Each organization can have up to 50, running as often as every 5 minutes.
- Runners: Your own machines run the agents Dispatch starts, in folders you approve.
- Temporary inboxes: Your agent makes a short-lived email address. Forward an email to it and the agent can read it; the address expires on its own.
- Temporary webhooks: Your agent makes a short-lived URL other services can send webhooks to, and reads each request as it arrives. The URL expires on its own.

## Guides

- [Documentation](https://help.bench47.com)
- [Connect an agent](https://help.bench47.com/humans/mcp/installing): Add Bench47 to Claude Code, Codex, Cursor and other MCP clients.
- [Team skills](https://help.bench47.com/humans/skills/overview): Publish your team's agent skills once; the mb CLI keeps every machine current.
- [Import skills from a repository](https://help.bench47.com/humans/skills/importing): Copy skills from a public Git repository into a team skill set from the dashboard.
- [Install mb](https://help.bench47.com/humans/skills/install-mb)
- [Dispatch](https://help.bench47.com/humans/dispatch/overview): Events from tools like GitHub, Sentry or Linear, mentions in a thread, emails, due dates on documents, and schedules start a coding agent on your own machines.
- [Schedules](https://help.bench47.com/humans/dispatch/schedules): Run an agent on a cron schedule in your time zone, with its prompt in a shared doc and every run in the job history.
- [Mentions](https://help.bench47.com/humans/dispatch/mentions): Mention Bench47 on a Linear issue, a GitHub issue or pull request, or in a Slack thread, and an agent answers in that thread.
- [Emails](https://help.bench47.com/humans/dispatch/emails): Give a lane an email address, and mail from an allowed sender starts a job with the message and its attachments.
- [Profiles](https://help.bench47.com/humans/dispatch/profiles): Say which agent, model, effort, sandbox and time limit a job runs with, and which machines may take it.
- [Job reports](https://help.bench47.com/humans/dispatch/job-reports): An agent ends each job by reporting that it finished, found nothing to do or needs something, and Bench47 decides what happens next.
- [Live activity](https://help.bench47.com/humans/dispatch/live): See an agent's messages and tool calls on the job's page while it runs and for 14 days after.
- [Encryption at rest](https://help.bench47.com/humans/dispatch/encryption): Which Dispatch content Bench47 stores encrypted, what stays readable, and how the keys work.
- [Set up a runner](https://help.bench47.com/humans/dispatch/runner)
- [Temporary webhooks](https://help.bench47.com/humans/mcp/webhooks): Your agent makes a short-lived webhook URL, registers it with a service or gives it to you to paste, and reads each request as it arrives.

## Comparisons

- [Bench47 vs Tessl](https://bench47.com/compare/tessl/): Tessl is built around its own skills registry. Bench47 uses skills from any registry, public or private, and adds the decisions and history your team builds up.
- [Bench47 vs Factory](https://bench47.com/compare/factory-ai/): Factory runs its own coding agent, Droid. Bench47 runs Claude Code, Codex, Cursor and the other agents your team already uses, all from shared context.
- [Bench47 vs 8090](https://bench47.com/compare/8090/): 8090 Software Factory asks your team to plan work in its own app. Bench47 builds a software factory into the tickets, branches and pull requests you use.
- [Bench47 vs your own setup](https://bench47.com/compare/build-your-own-software-factory/): You can build a software factory from instruction files, skills in Git and scripts. See what each part takes to run, and when buying one saves the effort.
- [Without a software factory](https://bench47.com/compare/coding-agents-without-a-software-factory/): Your team already uses Claude Code, Codex or Cursor. See what a software factory for teams changes when every agent works from the same context.

## Optional

- [Sitemap](https://bench47.com/sitemap-index.xml)
- [Blog sitemap](https://bench47.com/blog/sitemap-index.xml)