Skip to main content
This is optional. First-run is MCP setup: add the endpoint, finish the sign-in your client opens, then call proofable_context. To load the same agent into an editor project:
Same command for claude, codex, hermes, openclaw, or opencode. VS Code uses --apply cursor. This writes .proofable/mount.json and the host adapter. Hermes, OpenClaw, and OpenCode also write a pointer at .hermes/proofable.json, .openclaw/proofable.json, or .opencode/proofable.json. ACP jobs already attach hosted Proofable MCP; this pointer is project identity, not a second protocol. Start a new session so the project picks it up. The editor runs the tools. Proofable supplies identity, limits, and proofs. Cursor Cloud Agents can use the hosted MCP endpoint the same way. Proofable does not spawn those VMs for you.

​
From the agent card

Connect in copies:
Valid --apply values are cursor, claude, codex, hermes, openclaw, or opencode. VS Code uses --apply cursor. Run it in the project. If the host is not connected yet, run proofable setup first. The browser never writes host files.

​
Assistant tool

When signed in, call proofable_agent_mount with agentId. You can also use agentWallet or the identity proof ID in identityQHash. Suggested order:
  1. proofable_context
  2. proofable_agent_mount when acting as a specific agent
  3. proofable_verify_or_guide before sensitive actions; use proofable_proofs_check only for an explicit yes/no eligibility, gate, or access question

​
SDK

Use resolveRuntimeBundleFromMcp with your MCP transport, then applyRuntimeBundle(host, bundle, cwd) for project files. The same bundle shape is returned by the proofable_agent_mount MCP tool and the proofable mount CLI.

​
Where permissions are enforced

Your client controls its own tools. Proofable supplies verified context and permissions; your host remains responsible for enforcement.
Last modified on September 27, 2026