Agent product surface
MCP for Agents
Vibecodr exposes an agent-facing MCP product surface at `https://openai.vibecodr.space/mcp`. Connect it when you want an MCP client to find public context, help prepare launches, and request supported signed-in actions without handing the agent private platform authority.
Agent-facing MCP gateway and workspace guidance for safe Vibecodr automation.
Implementation focus
Use this guide when connecting an MCP client or designing agent workflows that should stay inside Vibecodr's public product contract.
Expected outcomes
- Connect agents to the right remote MCP endpoint and understand what the gateway is allowed to expose.
- Choose agent workflows that fit public discovery, launch prep, metadata help, and confirmed product actions.
- Know when to use native MCP tools, when Code Mode applies, and where MCP stops.
- Keep secrets, private drafts, storage internals, and deployment authority out of agent workflows.
What the MCP gateway is
The Vibecodr MCP gateway is the agent-facing product surface at https://openai.vibecodr.space/mcp. Connect it when you want an MCP client to understand Vibecodr context and help with supported product tasks.
Public tools can read public Vibecodr context. Signed-in tools use your authorized session and stay limited to the actions Vibecodr exposes for agents.
- Use MCP for public discovery, launch guidance, share copy, metadata help, and controlled authenticated actions.
- Do not use MCP to bypass the app, read private drafts, inspect secrets, operate storage directly, or receive deployment tokens.
- The default native surface is Streamable HTTP at /mcp with OAuth-compatible authentication for protected operations.
- Code Mode is separate and only applies when the MCP client or product flow explicitly enables it.
https://openai.vibecodr.space/mcpUseful agent workflows
MCP is a good fit for finding public examples, preparing a publish, checking runtime readiness, drafting share copy, and updating live metadata after confirmation.
Keep each step visible to the user. An agent should be able to explain whether it searched public context, prepared a publish, checked readiness, drafted copy, or requested a confirmed change.
- Ask agents to search public vibes, tags, social previews, and examples before starting a new idea.
- Use agent help to review dependencies, import status, preview readiness, publish steps, and launch notes before anything goes live.
- Draft titles, descriptions, changelog notes, and follow-up ideas before applying metadata changes.
- Public MCP tools do not hand agents private drafts, secrets, raw logs, provider account identifiers, or storage internals.
Where MCP stops
MCP can help coordinate Vibecodr tasks, but it does not replace the product surfaces that own your work. Put browser UI in a vibe, trusted backend behavior in a Pulse, and scheduled or webhook-driven work in automations.
For backend integrations, use the Vibecodr capability that matches the job: env.fetch for policy-aware outbound calls, env.secrets for server-side credentials, and MCP tools for confirmed agent workflows.
- Use https://openai.vibecodr.space/mcp for ordinary MCP clients.
- Signed-in actions still depend on your account access and permissions.
- Generated Code Mode code does not receive tokens, environment variables, or direct deployment wiring.
- If a workflow needs secrets, storage writes, schedules, or provider calls, model it as a Pulse or automation instead of browser code.
Example and read next
Example: an AI agent needs to publish and inspect Vibecodr work. Connect it to https://openai.vibecodr.space/mcp, use product-shaped native tools by default, and reserve Code Mode for explicitly enabled constrained execution.
Use these related pages when you need the next layer of guidance. They point to the most likely follow-up tasks, not every page that happens to touch the same system.
- Read next: Vibes & Pulses
- Read next: Pulse SDK & Handlers
- Read next: Automation Safety
- Read next: SEO & Discovery