ATP CORE · DEVELOPER TOOLS

Build with the open trust layer.

Use ATP alongside MCP, A2A, ANP, ADK, and your agent framework. Start with identity and add verifiable trust evidence as the implementations mature.

View ATP Core ↗Read the docs

GET STARTED

Choose your entry point.

Use Node.js 24 LTS. Copy an install command into your project’s terminal. For the CLI, start in a fresh directory. Libraries become part of your application when you import and call them.

01 / SDK

Add ATP to an existing agent

Add identity and trust utilities to your existing JavaScript or TypeScript agent. The standalone quickstart runs locally without an ATP account.

npm install atp-sdk@5.0.0
Runs locally in your terminal.

After installing: Import Agent and call Agent.quickstart in your code. Persist keys securely and integrate the services and tool boundaries your application needs.

SDK quickstart →

02 / CLI

Scaffold a new project

Create a folder, install the SDK, and open a local configuration screen. The screen saves settings; it does not mint keys or enforce policy.

npx create-atp-agent@1.6.0 my-agent
Runs locally in your terminal.

After installing: Stop the screen with Ctrl+C, enter my-agent, and run npm start. Edit agent.mjs to add your model, tools, and business logic.

CLI guide →

03 / Profiles

Use canonical profiles

Choose a built-in profile or define your own action gates. The evaluator returns allow, require_approval, or deny.

npm install atp-profiles@3.0.0
Runs locally in your terminal.

After installing: Call evaluateActionWithProfile before a tool runs. Your host must enforce the decision and its own command, path, network, and approval restrictions.

Explore profiles →

04 / Python

Gate tools in a Python agent

The Python port of the profile evaluator, with no dependencies. It returns the same decision and reason as the npm package for the same input, checked against shared conformance vectors.

pip install atp-profiles==3.0.0
Runs locally in your terminal.

After installing: Map each tool to an action category and call evaluate_action_with_profile before it runs. Block deny and require_approval unless your host has a real approval flow.

Python guide →

05 / Hermes

Guard Hermes Agent tool calls

A fail-closed pre_tool_call hook for Hermes Agent. It checks each matched tool against an ATP profile and the read and write paths you configure.

npm install atp-hermes-hook@0.1.0
Runs locally in your terminal.

After installing: Install it in its own folder, write atp-hermes.json, run the --check command, then add the hook to your Hermes config.yaml with absolute paths.

Hermes guide →

06 / DID

Work with did:atp

Construct canonical did:atp identifiers and key fingerprints from existing Ed25519 and ML-DSA-65 public keys.

npm install @atpdeveloper/did-atp@0.1.1
Runs locally in your terminal.

After installing: Call buildAtpV2Did with your domain, path, and public keys. Key generation, resolver hosting, credentials, and trust verification are separate integrations.

View source →

Run the available checks.

From an ATP Core checkout with dependencies installed, run npm run conformance. The suites cover available implementation behavior; passing them does not certify every Community Group draft capability.

Inspect the runner →

Read the draft specs.

ATP specifications are developed in the W3C Agent Trust Protocol Community Group. Community Group drafts are not endorsed by W3C and are not W3C Standards. Implementation support varies.

Explore specification drafts →

Explore working examples

See SDK, identity, and integration patterns with explicit notes about what is implemented.

Browse examples →