Skip to content

Repository files navigation

ACP (Agent Collaboration Protocol) Master Repository & Package Ecosystem

Status: Standards Track Specification: RFC-ACP-001..010 JSON Schema: Draft 2020--12 Reference Runtime: Python 3.13+ Tests: 18/18 Passing Release v1.0.0

Welcome to the official master repository for ACP (Agent Collaboration Protocol)—the open, cloud-native, transport-agnostic standard designed to serve as the distributed operating protocol and runtime fabric for autonomous AI agents (N:M mesh collaboration).


📦 Quick Installation & Package Ecosystem

We have built full distribution packages, Docker images, and SDKs so you don't have to manually manage source code:

1. Python Package (wheel & sdist)

Install the official Python reference runtime and CLI directly from our release artifacts or locally:

# Install pre-built Wheel directly from GitHub Release v1.0.0
pip install https://github.com/ZendanTeam/acp-protocol/releases/download/v1.0.0/acp_protocol-1.0.0-py3-none-any.whl

# Or install from source
git clone https://github.com/ZendanTeam/acp-protocol.git
cd acp-protocol
pip install .

Once installed, the acp command is available globally on your terminal!

2. TypeScript & Node.js SDK (@acp-protocol/sdk)

For frontend, Node.js, Next.js, or browser Web Worker agents:

cd sdk/typescript
npm install
npm run build

3. Docker & Docker Compose

Launch a complete sandboxed ACP Node (Gateway + Router + Memory Service + Dashboard) locally or in production:

docker compose up -d

🚀 Interactive CLI Commands (acp)

After installing via pip, you can run all 5 interactive protocol verification commands directly:

# 1. Inspect the live mesh topology and registered agent manifests
acp inspect-mesh

# 2. Execute a multi-agent DAG workflow with a Voting Gate (>= 66.7% consensus)
acp run-workflow

# 3. Simulate unrecoverable node failure and reverse Saga Rollback
acp simulate-saga-rollback

# 4. Demonstrate the 6-Tier Distributed Memory & CRDT LWW-Element-Set Sync
acp memory-demo

# 5. Simulate protocol-level economic SLA negotiation (PROPOSE -> COUNTER -> ACCEPT)
acp simulate-negotiation

🏗️ Automated CI/CD & GitHub Actions (.github/workflows/)

This repository is backed by complete, production-ready continuous integration and deployment pipelines:

  • ci.yml: Runs the 18-test pytest verification suite across Python 3.10, 3.11, 3.12, and 3.13, validates all 6 JSON Schemas (draft-2020-12), and builds the TypeScript SDK on every push and pull request.
  • release.yml: Automatically builds Python Wheels (.whl), builds and pushes multi-stage Docker container images (ghcr.io/ZendanTeam/acp-protocol:latest), and publishes release assets to GitHub Releases upon every version tag push (v*).

📚 Protocol Specification Directory (spec/)

  1. 01-architecture-and-vision.mdRFC-ACP-001: Vision, Core Principles, Mesh Topology, and Universal Envelope Canonicalization (Ed25519/HMAC-SHA256).
  2. 02-components-and-roles.mdRFC-ACP-002: Deep-dive specifications for all 15 components: Host, Agent, Gateway, Router, Registry, Discovery, Memory Service, Scheduler, Broker, Identity Provider (IdP), Marketplace, Runtime, Monitoring, Logging, and Telemetry.
  3. 03-communication-and-transport.mdRFC-ACP-003: Semantics for the 9 communication vectors (Agent→Agent, Human, Tool, Database, API, Browser, Robot, Cluster, Cloud), transport bindings (HTTP/1/2/3, QUIC, gRPC, WebSocket, Unix Socket, TCP, UDP, Bluetooth, mDNS, P2P Mesh), and native multi-modal streaming.
  4. 04-workflow-and-orchestration.mdRFC-ACP-004: DAG execution rules, task delegation, planning, SLA negotiation (PROPOSECOUNTERACCEPT), voting consensus gates (VOTING_GATE), exponential backoff retries, checkpoint snapshots, and reverse topological Saga Rollbacks (rollback_action).
  5. 05-distributed-memory.mdRFC-ACP-005: Unified 6-tier distributed memory (Short-Term LRU, Long-Term KV, Shared CRDT LWW-Element-Set, Vector Cosine Sim, Knowledge Graph RDF/LPG Triples, Encrypted Enclaves), replication loops, and copy-on-write snapshots.
  6. 06-security-and-zero-trust.mdRFC-ACP-006: Zero-Trust request lifecycle, OAuth2/OIDC token exchange (RFC 8693), mutual TLS (mTLS 1.3/QUIC), cryptographic capability tokens, ABAC claims, replay eviction caches (message_id + nonce), token bucket rate limiting (1,000 fps), and the automated Secrets Vault.
  7. 07-discovery-and-marketplace.mdRFC-ACP-007: Multi-tier discovery (mDNS + SWIM Gossip + Registry Queries), semantic vector skill matching, and the ACP Marketplace catalog for verified Skills, Plugins, Tools, Models, Workflows, Templates, and Knowledge Packs.
  8. 08-runtime-and-observability.mdRFC-ACP-008: Runtime targets across Docker, K8s, Serverless (Lambda/Workers), Bare Metal (cgroups v2), Edge/IoT (Micro-ROS), Mobile/Desktop, and Browser (Wasm Web Workers), and OpenTelemetry (traceparent: 00-4bf92f35...-01) metrics and circuit breakers.
  9. 09-sdk-specifications.mdRFC-ACP-009: Architectural layers, client abstractions, and idiomatic code signatures across 14 programming languages (Go, Rust, C#, Java, Kotlin, Swift, Python, JavaScript, TypeScript, PHP, Ruby, Lua, C, C++).
  10. 10-developer-experience-and-migration.mdRFC-ACP-010: CLI (acp), GUI Dashboard, Visual Studio Workflow Builder, step debugger, chaos simulation (acp-test), code generator (acp codegen), and complete migration guides from MCP, LangGraph, CrewAI, and AutoGPT.

📋 Formal JSON Schemas (schemas/)

All wire payloads are formalized using draft-2020-12 JSON Schemas:

  • envelope.schema.json — Universal ACP Frame Envelope (routing headers, W3C traceparent, signatures, capability tokens, nonces, payload bodies).
  • message.schema.json — Payloads for REQUEST, RESPONSE, STREAM_CHUNK, NEGOTIATION_PROPOSAL, VOTE_CAST, and TASK_DELEGATION.
  • agent-manifest.schema.json — Manifest published to Registry (did:acp:..., skills, endpoints, compute requirements, security profiles).
  • workflow-dag.schema.json — Directed Acyclic Graph specification, nodes, dependencies, voting consensus gates, and rollback_action declarations.
  • capability-token.schema.json — Cryptographically signed zero-trust token granting time-bound action rights over resource URI patterns.
  • memory-operation.schema.json — Operations for reading, writing, vector querying (min_similarity), graph querying (max_depth), CRDT syncing, and snapshotting across the 6 memory tiers.

About

No description, website, or topics provided.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages