Skip to content

Native AI connection In development

It speaks to AI. Natively.

An MCP server so Claude, ChatGPT, Cursor, VS Code or your own agents can search, verify and compare records without anyone writing a line of integration.

An AI can make up an answer. Not a proof.

Either the fingerprint matches or it does not. Your agents cite sealed facts, with their verification link, instead of guessing.

Connect

Add connector. Authorise. Done.

A remote server at api.knowyourparcel.com/mcp over HTTP transport. Authorisation uses OAuth 2.1 with PKCE and dynamic client registration: what assistants expect when you press “add connector”.

In the dashboard you choose which account and which permissions the AI sees, and you revoke it whenever you like. For your own agents, an API key in the header works too.

{
  "mcpServers": {
    "kyp": {
      "type": "http",
      "url": "https://api.knowyourparcel.com/mcp"
    }
  }
}
{
  "mcpServers": {
    "kyp": {
      "command": "npx",
      "args": ["-y", "@knowyourparcel/mcp"],
      "env": { "KYP_API_KEY": "kyp_test_…", "KYP_ENV": "sandbox" }
    }
  }
}

Locally, too. The @knowyourparcel/mcp package runs on your machine, pointing at production, the test environment or your local sandbox.

Tools

Twelve tools. Ten only read.

  • kyp_find_recordsFinds records by order, tracking, customer, station or date.
  • kyp_get_recordReturns the record: size, weight, images by temporary link, label and seal.
  • kyp_verifyChecks that a record or a file is still intact.
  • kyp_compareSets two records side by side, such as outbound and return.
  • kyp_check_declarationCross-checks what was declared against what was measured.
  • kyp_get_timelineThe parcel passport: every hand, every event, every record.
  • kyp_get_riskRisk score with its signals and the rule behind each one.
  • kyp_transit_timesReal transit times by carrier and route, with their sample.
  • kyp_device_statusYour stations’ status: online, calibration, incidents.
  • kyp_usageThis month’s usage against what your licence includes.
  • kyp_build_dossierBuilds the claim file. It creates a document; it never touches the evidence.writes
  • kyp_create_journeyPublishes a parcel page. Asks for confirmation first.writes

With the Vault and Analytics come kyp_vault_manifest (the manifest only, never the files) and kyp_journey_stats.

Resources

  • kyp://records/{ref}

    The record as signed JSON, ready to cite.

  • kyp://docs/…

    The documentation guides, so the agent can teach itself.

Prompts

  • Review a claim

  • Compare outbound and return

  • Daily device summary

Try it with your data, from the dashboard.

A sample chat calls the tools with your account so you can see what an AI would answer before connecting yours.

Operator

The customer says order ORD-88213 arrived broken. How did it leave?

Know Your Parcel

kyp_find_records · reference: "ORD-88213" → 1 outbound record

Know Your Parcel

kyp_verify · KYP-2026-000123 → intact · attested capture

Assistant

It left on 16 October at 09:41 UTC: 32.0 × 24.5 × 18.0 cm and 2.340 kg, with no visible packaging damage in either image. The record is intact. Shall I prepare the file for the carrier?

Operator

Yes, in Spanish.

Know Your Parcel

kyp_build_dossier · claim: damage, lang: es, confirm: true → ready

Assistant

Claim file ready, with a public verification link. Attaching it to the claim.

Rules

The AI sees just enough. And everything is written down.

  1. 01

    Read-only by default

    The two tools that write stay off until an admin turns them on.

  2. 02

    Confirmation to write

    Building a claim file or publishing a page requires explicit confirmation in the call.

  3. 03

    Every call, in the log

    With the name of the assistant that made it. You see it in your API log and in the account audit trail.

  4. 04

    Your account only

    Tools return data from the authorised account and nothing else. Images by temporary link, never raw.

  5. 05

    Their own limits

    A per-minute cap for agents, separate from your integration’s.

Documentation machines can read.

  • llms.txt and llms-full.txt

  • OpenAPI 3.1 ready for function calling

  • Tool descriptions with examples, and errors that explain what to do

  • A “For agents” page with configuration ready to copy

Want to connect your AI agent during the pilot?

Request access