MCP Manifest
Generator

Turn any API into a production-ready MCP server in minutes. With AI-readability optimization, safety scoping, and a test harness built in — so you ship integrations Claude can actually use.

Step 1. Input

Paste an OpenAPI spec or Postman collection, or try a sample API.

Try a sample
Step 1 of 5 · Parse

Step 2. Review descriptions

The optimizer rewrote your tool descriptions for AI-readability. Accept what helps; keep what you had.

0 tools
+0pp avg lift
0 tools have recommended rewrites that beat the baseline.
You can accept all and override individually below.
Step 2 of 5 · Optimize

Step 3. Safety scoping

The classifier flagged destructive operations. These require approval before Claude can run them.

0 require approval
0 read-only
Step 3 of 5 · Classify

Step 4. Test harness

50 simulated prompts per tool, run in a sandbox. Catches malformed responses before they ship.

Ready
—

The harness will run 50 simulated prompts against each tool in your manifest, using Claude Haiku 4.5 in a sandboxed environment. Failed calls and malformed responses are surfaced per-tool before the manifest is published.

Step 4 of 5 · Test

Step 5. Ship

Download your validated manifest, or copy the registry command.

Signed 0 tools

Your manifest is ready. Commit it to your repo and teammates' Claude Code will auto-install on git pull. Or push to the public registry for community use.

How it works

Parse — Extracts every operation from your OpenAPI spec or Postman collection into a normalized tool list.
Optimize — Rewrites tool descriptions for AI-readability. Selection accuracy climbs from 71% baseline to ≥92% on optimized manifests.
Classify — Flags every destructive operation (deletes, writes, charges, sends) with rationale and example invocation. Requires explicit override.
Test — Runs 50 simulated prompts per tool in a sandboxed harness. Catches malformed responses before they reach production.
Ship — Export the signed manifest or push to the public registry. Teammates auto-install on git pull.