Use passo.build inside your coding agent
This mode is for people who want their coding agent to drive the production readiness flow without constantly switching to a web console.
The agent still writes and edits code. passo.build gives it a process:
- scan the local repo;
- understand what product is being built;
- ask missing questions in the agent chat;
- produce a concrete implementation plan;
- track evidence and locks so multiple agents do not overwrite each other;
- compare deploy options;
- require confirmation before anything spends money or changes cloud resources.
This is useful for Claude Code, Codex, Cursor, Antigravity, Lovable/Base44-style builder agents, and any other host that can talk to a remote MCP server.
What the user experiences
The user should see normal agent-native prompts, not a wall of protocol details:
- "I found a Next.js frontend and no durable database. Does this product store user data?"
- "You selected file uploads. I will plan object storage plus database metadata."
- "These three security issues block deploy."
- "DigitalOcean is cheapest for this shape; AWS costs more for the same starter resources."
- "Deploy is ready. Confirm the region, spend cap, and account mode."
The agent can show those decisions in its own UI, while passo.build stores the project state and evidence.
What the agent sends
The agent reads code locally and sends structured findings, not a raw source-code dump. The important payload is a product assessment:
- frameworks and runtimes detected;
- product type and missing answers;
- database, file, auth, job, API, and integration signals;
- security and deployment gaps;
- the implementation plan it wants the user or agent to execute.
Current rollout status
| Status | What it means |
|---|---|
| Live now | The web console is the stable UI at passo.build. |
| Rolling out | The MCP/coding-agent flow is the agent-native interface. Verify the live tool catalog with tools/list before relying on a specific tool. |
| Target UX | Local login/session binding from the agent host. Users should not have to paste raw tokens into random files. |
| Fallback | If the MCP endpoint or host auth is unavailable, use the web console for the same project state. |
What the agent can do automatically
The agent can read the approved local repo, summarize signals, ask questions, draft implementation tasks, write code in a worktree, run tests, and attach evidence.
The agent must ask before deploy, cloud mutation, billing, SSO changes, secret handling, or accepting high-risk security findings.