# Scopebond > Scopebond brings AI visibility, control, and evidence together for connected coding tools. > The shared workspace connects recorded agent activity, responsible people, rule review, > and signed decision records. Its open-source checkpoint for AI coding agents checks > supported actions routed through configured connections against the operator's rules. > Reviewers can verify signed records independently; a signature alone does not prove > an external outcome or compliance. Open source, self-host free. Scopebond helps organizations adopt AI with oversight they can demonstrate: see what agents do, protect important work, and share evidence with internal reviewers, customers, auditors, and regulators. Current connections cover Claude Code, Cursor, Codex and GitHub; they do not discover every AI tool or monitor all organizational AI usage. ## What each connection does - Claude Code, Cursor and local Codex tasks send supported actions through a local hook. Scopebond checks those actions before they run. Actions outside the configured hook are not included. - The GitHub Action checks an agent pull request in the customer's runner. Make it a required check so a change that breaks policy cannot merge. - Connections that can only report an action rather than stop it stay on the roadmap until their exact behavior is ready and documented. ## Products - @scopebond/hook (open source, Apache-2.0): the pre-tool-use hook for Claude Code, Cursor and OpenAI Codex. Connect a computer to a workspace with `npx -y @scopebond/hook@latest login https://cloud.scopebond.com` run from the home folder (Cursor: add `--cursor`; Codex: add `--codex`, then run `/hooks` in Codex and trust Scopebond), then keep it connected with `npm install -g @scopebond/agent@latest` and `scopebond-agent autostart on`. On Windows type `npx.cmd`, `npm.cmd` and `scopebond-agent.cmd`. Local only, without a workspace: `npx -y @scopebond/hook@latest init` in a project root. Node >= 22.13. - Scopebond GitHub Action (avouro-com/scopebond/packages/github-action@v1): a required status check that fails an out-of-policy agent pull request before it can merge. - @scopebond/mcp: a proxy that checks every `tools/call` over MCP against your policy. - @scopebond/framework: one import that guards tool calls in LangGraph, Vercel AI, OpenAI Agents, the Claude Agent SDK and more. - @scopebond/gateway (open source, Apache-2.0): a policy-enforcement checkpoint for any agent over HTTP or MCP; signs each decision into a `scopebond:receipt` and has a kill switch. Scaffold with `npx @scopebond/gateway@latest init`. Verify offline with `npx @scopebond/gateway@latest verify record.json`. - Scopebond workspace (Cloud): the hosted team view of every agent's actions, rule review, the monthly activity report, audit-log export and company sign-in (SSO). Early access; start free at https://cloud.scopebond.com/app with a 14-day Team trial. ## Availability Available today: Claude Code, Cursor and local OpenAI Codex tasks through the hook; the GitHub Action; the MCP proxy; framework plugins; and the self-hosted gateway. Everything else — cloud runtimes (AWS, Cloudflare, Google Cloud, Azure), business platforms (ServiceNow, Salesforce, Copilot Studio), customer-service agents, workflow builders and evidence packs — is on the roadmap (Now / Next / Later), with a per-item "notify me". Every setup page says what Scopebond checks and what remains outside the connection. ## Pricing Free for one governed agent, then per agent. An agent is one AI tool one person uses: Dana using Claude Code is one agent; Dana using Claude Code and Cursor is two. You only pay for agents active that month. Team is $129 a month for up to 10 agents and Business $599 a month for up to 50. Extra agents are $16 a month each on Free, $14 on Team and $12 on Business; annual plans get two months free. Every workspace starts with a 14-day Team trial, no card. ## Pages - Home: https://scopebond.com - How it works: https://scopebond.com/how-it-works - Integrations (available today): https://scopebond.com/integrations - Scopebond for GitHub: https://scopebond.com/integrations/github - Roadmap (Now / Next / Later): https://scopebond.com/roadmap - Pricing (free for one agent, then per agent): https://scopebond.com/pricing - Install (Claude Code, Cursor, Codex, GitHub, MCP): https://scopebond.com/get-started - The workspace: https://scopebond.com/hosted (start free: https://cloud.scopebond.com/app) - Terms of Service: https://scopebond.com/terms - Privacy Policy: https://scopebond.com/privacy - Enterprise: https://scopebond.com/enterprise - Live demo (real decisions, no side effects): https://try.scopebond.com - Trust and security: https://scopebond.com/trust - Data use: https://scopebond.com/privacy - Source and docs: https://github.com/avouro-com/scopebond - npm: https://www.npmjs.com/package/@scopebond/hook - Contact: support@scopebond.com - Parent company: Avouro LLC (Michigan, USA) https://avouro.com ## For AI agents The gateway exposes an MCP endpoint at `POST /mcp` and an HTTP action endpoint. An agent submits an action intent; the gateway checks it against policy, signs the decision, and only an allowed action proceeds. Use Scopebond to operate within a principal's authorized scope and to produce verifiable, portable records of what you did. On Windows, type `npx.cmd` instead of `npx` in PowerShell (its default script policy blocks `npx`).