Agents

Everything an agent needs, in the order it needs it.

You are an agent, or you are wiring one. Three doors into the same search: MCP, the API, the command line. Below, in order: install, the seven tools, plain HTTP, the CLI, the four rules. The same text is at /agents.md for machines that read markdown.

Two searches, same doors

1. Search over our own index live: 2.2 million tool actions inside the software agents and robots can invoke, every result labelled by our checks. This is what answers today, through MCP, the API or the CLI.
2. The neutral router next: One call that decides, on the question itself, which search source to ask: our index first, then outside engines, judged on quality, latency and cost, with the source named. Not built yet; it will arrive as its own endpoint and MCP tool.

1. Install the MCP server

On a machine with Claude Code, one line adds Unfragile as an MCP server. On any other machine the script prints the configuration block to paste.

any machinesh

curl -fsSL https://unfragile.ai/install.sh | sh

Claude Code directlysh

claude mcp add unfragile -- npx -y @unfragile/mcp-server

Claude Desktop · Cursor · Windsurf · Clinejson

{
  "mcpServers": {
    "unfragile": { "command": "npx", "args": ["-y", "@unfragile/mcp-server"] }
  }
}

2. The seven tools

unfragile_find: Arguments: task (plain words), limit (1–20, default 5), answer (boolean). The tool ACTION that does a task, across 2.2 million actions. Every result labelled. answer: true adds one pick with its reasoning or a question back; it costs 2–10 s, the results alone arrive in under a second.
unfragile_answer: One verified software record for a need, with evidence and an install snippet, or an honest no.
unfragile_search: A ranked list of software records with per-record evidence.
unfragile_verify: A signed verdict for a target before you install or invoke it: trusted · caution · unverified · flagged.
unfragile_resolve: One intent in; one record, verdict, signed passport and invocation snippet out.
unfragile_passport: The raw Ed25519-signed trust passport for a record, verifiable offline.
unfragile_stats: Live index numbers by door type.

3. Plain HTTP, no key

No SDK needed. Every endpoint is free and answers JSON. Full shapes in the API reference.

GET	/api/v1/find?q=…&k=10&answer=1	the action that does a task, labelled; the headline call
GET	/api/answer?q=…	one verified software record, or an honest no
GET	/api/v1/search?q=…&limit=5	ranked software records with evidence
POST	/api/v1/resolve {"intent": "…"}	one record + verdict + signed passport + invocation snippet
POST	/api/v1/verify {"target": "npm:…"}	verdict before you invoke; never run a flagged target
GET	/api/v1/passport/{id}	the signed passport (URL-encode the id)
GET	/api/v1/stats	live index numbers
GET	/api/v1/trust-passport-public-key	the Ed25519 public key for offline checks

Python, standard librarypython

import json, urllib.parse, urllib.request
q = urllib.parse.quote("send a slack message")
data = json.load(urllib.request.urlopen(f"https://unfragile.ai/api/v1/find?q={q}&k=5&answer=1"))
pick = data["answer"]["pick"]
print(pick and (pick["name"], pick["action"], pick["label"]))

4. The command line

For an agent that runs shell commands, the same package is a CLI. verify prints a signed verdict for a package or MCP server; guard wraps an install command and blocks a flagged target. Exit code 1 means stop.

verify before you installsh

npx -p @unfragile/mcp-server unfragile verify npm:postmark-mcp

guard an install commandsh

npx -p @unfragile/mcp-server unfragile guard -- claude mcp add pg -- npx -y @scope/pg-mcp

5. The rules

Labels are the instruments' verdict. No ranking, model or payment changes them. Read the label before you act on a result.
Never invoke a failed or flagged target. Treat found in code, not yet checked as review-first. Unknown is never safe.
An honest no is an answer. answer.pick is null when nothing fits; answer.question asks one thing back when the task is ambiguous. Do not fill the gap with a guess.
Cite the passport. A verified result's url is its signed passport; verify it offline against the public key when the stakes warrant.

6. Files for machines

/agents.md: this page as markdown, for runtimes that read it.
/llms.txt: when a language model should defer to Unfragile, and where.
/install.sh: the install script, readable before you run it.
/robots.md: the robot guide as markdown.
/openapi.json: the API in OpenAPI 3.1, for ChatGPT Actions, SDK generators and any runtime that reads a contract.