Make your software speak agent
Any software → agent-, machine-, and LLM-ready.
Point Unfragile at your API, repo, or docs. We decompose it into discrete capabilities and hand back the artifacts that let any agent or LLM discover, trust, and invoke it — a signed agent.txt manifest and unfragile.yml capability schema today; a one-command MCP server and CLI wrapper next.
The on-ramp from human software to the agent economy — and your listing on the graph the agents resolve through. Free for open source.
What you get
Pass an OpenAPI/Swagger URL and all three are generated live — agent.txt, a working MCP server, and a CLI wrapper. A GitHub repo URL returns the inferred manifest. Snippets below are illustrative.
# https://yourcompany.com/agent.txt
name: yourcompany
version: 1.2.0
api: https://api.yourcompany.com
capabilities:
- id: invoices.get
method: GET
path: /v1/invoices
- id: invoices.post
method: POST
path: /v1/invoices// src/index.ts — @modelcontextprotocol/sdk + zod
const server = new McpServer({ name: "yourcompany", version: "1.2.0" });
server.tool("invoices_get",
"Get invoice (GET /v1/invoices/{id})",
{ "id": z.string(), query: z.record(z.string(), z.any()).optional() },
async (args) => { /* calls your live API */ },
);
// …one tool per operation, ready to npm install && build.# zero-dependency Node CLI — one command per operation export YOURCOMPANY_API_URL=https://api.yourcompany.com export YOURCOMPANY_API_KEY=... # optional bearer token node bin/cli.mjs invoices_get --id inv_123 node bin/cli.mjs --help
No OpenAPI spec?
The spec path generates a working MCP server + CLI right now. Full codegen straight from a bare GitHub repo (README → wired endpoints) is rolling out — join the waitlist and we'll email you when your repo can generate code, not just a manifest.
How it works
- step 1Paste
Give us your OpenAPI / Swagger JSON URL.
- step 2Generate agent.txt
We parse every operation into a capability and hand back a ready-to-host agent.txt manifest — instantly.
- step 3Generate the code
With an OpenAPI/Swagger URL you also get a working MCP server and a CLI wrapper — one tool per operation. Host the manifest at /agent.txt and ship.
Register & sign — so agents trust you
Agents now verify before they invoke. Register your tool to declare its scope and get a signed attestation + a hostable unfragile.yml. Then prove ownership (a 6-digit code) so your verdict becomes trusted — registration alone never inflates trust.
curl -X POST https://unfragile.ai/api/v1/register \
-H "Content-Type: application/json" \
-d '{
"target": "npm:@yourco/mcp",
"declaredPermissions": ["data.read"],
"contactEmail": "dev@yourco.com"
}'Returns a signed registration, your unfragile.yml, and the claim link. Then verify it anytime: POST /api/v1/verify {"target":"npm:@yourco/mcp"}.
Prefer the command line? The response includes your agent.txt, MCP server, and CLI:
curl -X POST https://unfragile.ai/api/v1/build \
-H "Content-Type: application/json" \
-d '{"url":"https://api.yourcompany.com/openapi.json"}'Codegen is free. Already have an MCP? List it at /submit · want agents to find it? see distribution · read the resolver spec