Talocode Loop

Search a public page, verify a claim against that page, and keep the receipt.

The same call is a CLI, an SDK, a local MCP server, and a hosted API. A failing verify stays a failed receipt. It does not pretend the run succeeded.

A hosted credit moves only when a source loads. Checking more claims on that same page does not add a credit. A timeout or an HTTP error costs nothing. The local CLI does not use credits.

Install

npm install -g @talocode/loop

The command is talocode-loop. From this repo, node bin/loop.mjs does the same thing.

CLI

talocode-loop run \
  --query="Talocode official mint" \
  --url=https://talocode.site/llms.txt \
  --claim=6ptxwABxQz8zMhwhiPeVgRgWjGMdVcEBFBv8v8C3ory

Research brief. Fetch every source you name. recommendation is do_not_ship when any source fails, and review when every source responds. Page text is not printed. This is not a decision to ship.

talocode-loop brief \
  --question="Is this source enough?" \
  --url=https://talocode.site/llms.txt

Audit checks up to 8 claims against one page. Repeat --claim for each one. Locally this spends nothing.

talocode-loop audit \
  --query="Talocode token facts" \
  --url=https://talocode.site/llms.txt \
  --claim=6ptxwABxQz8zMhwhiPeVgRgWjGMdVcEBFBv8v8C3ory \
  --claim="500,000 TCODE fixed" \
  --claim="unlimited supply"

Local server, no credits:

talocode-loop serve --port=8787

GET /v1/loop/health, POST /v1/loop/run, and POST /v1/loop/audit.

SDK

import { TalocodeLoop } from "@talocode/loop";

const loop = new TalocodeLoop();
await loop.run({
  query: "Talocode official mint",
  url: "https://talocode.site/llms.txt",
  claim: "6ptxwABxQz8zMhwhiPeVgRgWjGMdVcEBFBv8v8C3ory",
});

await loop.brief({
  question: "Is this source enough?",
  urls: ["https://talocode.site/llms.txt"],
});

await loop.audit({
  query: "Talocode token facts",
  url: "https://talocode.site/llms.txt",
  claims: [
    "6ptxwABxQz8zMhwhiPeVgRgWjGMdVcEBFBv8v8C3ory",
    "500,000 TCODE fixed",
    "unlimited supply",
  ],
});

Leave the base URL empty to run on this machine. Set it to spend cloud credits:

const hosted = new TalocodeLoop({
  baseUrl: process.env.TALOCODE_LOOP_URL || "https://api.talocode.site",
  apiKey: process.env.TALOCODE_API_KEY,
});

MCP

talocode-loop mcp

Tools: loop_run, loop_brief, and loop_audit.

The hosted MCP is https://api.talocode.site/mcp with the same API key. Hosted tools:

Tool	Credits
loop_health	free
loop_pricing	free
loop_capabilities	free
loop_run	1 if the page loads
loop_brief	1 per page that loads, maximum 5
loop_audit	1 if the page loads, up to 8 claims

Hosted API

Base URL: https://api.talocode.site

Method	Path	Auth	Credits
GET	/v1/loop/health	no	0
GET	/v1/loop/pricing	no	0
GET	/v1/loop/capabilities	no	0
POST	/v1/loop/run	Bearer key	1 if the page loads
POST	/v1/loop/brief	Bearer key	1 per page that loads
POST	/v1/loop/audit	Bearer key	1 if the page loads, up to 8 claims

curl -s https://api.talocode.site/v1/loop/run \
  -H "authorization: Bearer $TALOCODE_API_KEY" \
  -H "content-type: application/json" \
  -d '{"query":"Talocode official mint","url":"https://talocode.site/llms.txt","claim":"6ptxwABxQz8zMhwhiPeVgRgWjGMdVcEBFBv8v8C3ory"}'

The hosted route accepts public https URLs only. It refuses localhost, private addresses, and link-local addresses. The receipt returns the status, up to eight matching lines, whether each claim was present, and the line that contained it. It does not return the page. Eight claims on one page are still one credit.

usage.credits is 0 when nothing loaded.

What this does not do

It does not browse a logged-in site. review means every named source responded. It does not mean ship.