Originally published on Truthlocks Blog
When we set out to build machine identity infrastructure, we faced a choice that would shape everything that followed. We could build a proprietary system that locks customers into our platform, or we could create an open protocol that anyone can implement and let our platform compete on execution quality.
We chose open. The Machine Agent Identity Protocol (MAIP) is published under the Apache 2.0 license at github.com/truthlocks/maip. The specification has zero proprietary dependencies. Any organization can implement a MAIP compatible system without paying us a cent or asking our permission.
This was not an altruistic decision. It was a strategic one. Machine identity will only work if it becomes a shared standard. A world where every vendor has its own incompatible agent identity system is worse than no agent identity at all, because it fragments trust instead of building it. We believe the best way to establish the standard is to make it free, open, and excellent.
What the Protocol Covers
MAIP is a complete specification for the agent identity lifecycle. It is not a framework, not an SDK, and not a product. It is a protocol: a set of rules that any system can implement to be interoperable with any other MAIP compatible system.
The protocol defines six core areas:
Agent identity and registration. How agents are assigned cryptographically verifiable identities using Decentralized Identifiers (DIDs). The DID method, key generation requirements, registration workflow, and identity metadata schema are all specified. Any system that follows the spec can issue and verify MAIP agent identities.
Scope based authorization. How agent permissions are expressed and enforced using a hierarchical scope system. The scope syntax, wildcard and negation semantics, scope validation rules, and authorization decision flow are specified. Any system that follows the spec can make consistent authorization decisions for MAIP agents.
Trust score computation. How behavioral signals are combined into a composite trust score. The five trust factors (behavioral compliance, scope adherence, anomaly score, peer attestations, session hygiene), their weights, the scoring algorithm, and the update cadence are all specified. Any system that follows the spec will compute compatible trust scores.
Session management. How bounded execution contexts are created, validated, and revoked. The session token format (JWT based), required claims, expiry semantics, scope ceiling enforcement, and revocation mechanism are specified. Any system that follows the spec can issue and validate MAIP session tokens.
Delegation credentials. How agents delegate authority across trust boundaries. The delegation credential format, scope limitation rules, temporal constraints, chain depth limits, and verification procedure are specified. Any system that follows the spec can issue and verify MAIP delegation credentials.
Witness events. How significant agent actions are recorded for auditability. The event schema, required fields, cryptographic chaining requirements, and inclusion proof format are specified. Any system that follows the spec can produce and verify MAIP witness events.
Design Principles
We designed MAIP with three principles that guided every decision.
Implementable. Every section of the spec includes enough detail that a competent engineering team can build a compliant implementation without contacting us. If a section is ambiguous, that is a bug in the spec, and we want to hear about it.
Minimal. The protocol specifies the minimum necessary for interoperability. It does not dictate internal implementation details, database schemas, or deployment architectures. You can build a MAIP compatible system on PostgreSQL, on DynamoDB, on a blockchain, or on flat files. The protocol does not care as long as the external behavior matches the spec.
Extensible. The data models include extension points for additional metadata, custom trust factors, and domain specific fields. These extensions are namespaced to avoid collisions between implementations.
SDKs and Integration
Alongside the protocol specification, we publish reference SDKs in three languages:
The JavaScript/TypeScript SDK (@truthlocks/maip-sdk on npm) provides client classes for agent registration, session management, scope validation, and trust score queries. It works in Node.js and in the browser.
The Python SDK (truthlocks-maip on PyPI) provides the same capabilities with native async support and full type annotations. It is designed for AI/ML teams working with frameworks like LangChain, CrewAI, and AutoGen.
The Go SDK (github.com/truthlocks/maip-sdk-go) provides the same capabilities with idiomatic Go interfaces. It is designed for infrastructure teams building high performance services.
All three SDKs are also Apache 2.0 licensed. They are reference implementations, not the only way to interact with MAIP. If you prefer to call the API directly or build your own client library, the protocol spec is your guide.
IDE and CI/CD Integration
For developer workflow integration, we publish a VS Code extension that adds MAIP identity context to your editor, a JetBrains plugin for IntelliJ based IDEs, a GitHub Action that validates agent identities in your CI/CD pipeline, and a CLI tool (@truthlocks/protect-cli) for local development and testing.
These tools make it easy to adopt MAIP incrementally. You can start by adding the GitHub Action to your pipeline to verify that agents interacting with your repositories have valid MAIP identities, then expand to runtime enforcement as your infrastructure matures.
Contributing
The protocol is maintained in the open. We welcome issues, pull requests, and RFC style proposals for protocol changes. Major changes go through a review process where community feedback is incorporated before the spec is updated.
If you are implementing MAIP and find a gap, ambiguity, or improvement opportunity, open an issue. If you are building a MAIP compatible system and want to contribute interoperability tests, we will work with you to add them to the test suite.
Machine identity is too important to be owned by any single company. MAIP is the protocol. Truthlocks is one implementation. We want there to be many.
Truthlocks provides machine identity infrastructure for AI agents. Register, verify, and manage non-human identities with trust scoring and instant revocation.
Top comments (0)