PromptO

Documentation

PromptO is the visual runtime for MCP: it sits in front of any MCP server and gives its tools polished in-client interfaces, an honest text fallback for hosts without app rendering, and guardrails (budgets, busy signals, confirmation gates) that upstreams do not have to implement themselves.

Connect from a host (hosted)

Each tenant gets streamable HTTP MCP endpoints, one per configured upstream server:

https://<this host>/mcp/{tenant}/{server}
Authorization: Bearer <tenant key>

Hosts that cannot send headers can embed the key in the path: /mcp/{tenant}/{server}/k/{key}. Add the URL as a custom connector in your host application (Claude, ChatGPT developer mode, or any MCP client that speaks streamable HTTP).

Run it yourself (stdio)

The single binary wraps any local MCP server described by a manifest:

prompto stdio --manifest prompto.manifest.json

The manifest names the upstream command, binds tools to experiences (views), and declares per-tool policies: call budgets, concurrency, confirmation gates, annotations, and fallback formatting. No upstream code changes are required.

What hosts see

• MCP Apps hosts render the bound experience as an interactive panel (askflow, auto data views).
• Text-only hosts get a schema-driven Markdown rendering of the same structured result, and an honest notice that the panel was skipped, stamped with a support id.
• A2UI-capable clients additionally receive the result as a declarative A2UI v0.9.1 surface.
• Sampling and elicitation from upstream servers are bridged through to the host rather than blocked.

Support ids

Every session gets a short support id, shown in fallback notices. Quote it when asking for help; it lets the operator look up what your host claimed it could render. See the support page.