Invocation guide

How to make an app invocable by AI agents

Being found answers what can help. Being invocable answers whether an agent can call a real capability safely and correctly. They are separate systems with separate tests.

01

Choose an interface that matches the product

A web app does not become agent-invocable merely by adding an llms.txt file. Invocation requires an actual interface: an MCP server, a documented API, an agent card for a compatible agent, or another client-supported mechanism.

Do not advertise a transport until it is live. Deploy to Agents can audit public surfaces and exposes its own remote MCP endpoint; it does not claim to convert an arbitrary API into a working MCP server automatically.

02

Describe tools for correct selection

Tool names and descriptions should identify the user job, required inputs, meaningful output, side effects, and constraints. Schemas should reject ambiguous input instead of relying on prose. Read-only and destructive operations should be distinguishable.

Keep the public listing, live tool inventory, documentation, and repository aligned. A directory description cannot repair a stale schema, and a healthy handshake does not prove completion of a useful task.

03

Test invocation as a sequence

Verify transport negotiation, initialization, tool enumeration, schema validation, authorization boundaries, execution, and output interpretation. Record each stage so a retrieval success is not reported as task success.

A strong invocation receipt names the client, endpoint, tool, sanitized inputs, result class, time, and failure boundary. Production claims should come from those receipts, not from an MCP badge.

Verification layer

Primary sources

These links support the definitions and product facts above. They do not constitute independent endorsement.

Primary source ↗

Deploy to Agents public product facts

Machine-readable product facts, public surfaces, limitations, and verification links.

https://deploytoagents.com/llms.txt
Primary source ↗

Official MCP Registry: About

Primary documentation for the Registry's public metadata and downstream discovery role.

https://modelcontextprotocol.io/registry/about
Primary source ↗

Official MCP Registry publishing quickstart

Primary instructions for server.json publication and API verification.

https://modelcontextprotocol.io/registry/quickstart
Measure before you claim

See what agents can observe today.

Start from one public URL. Keep technical readiness, retrieval, recommendation, and invocation as separate evidence.

Create your profile