OPEN PROTOCOL — APACHE 2.0

The missing trust layer for autonomous agents

MCP connects agents to tools. A2A connects agents to agents. Nobody verifies that what they find is safe, compatible, or trustworthy. ASPL does. And it absorbs both.

8Message Types
6Security Layers
32Tests Passing
2Adapters (MCP+A2A)
POSITION IN THE STACK

We don't replace MCP or A2A. We sit above them.

ASPL is the trust and supply layer. MCP handles tool-calling. A2A handles agent communication. ASPL handles the question they both leave open: can I trust this?

ApplicationLangChain, CrewAI, AutoGen
ASPLfind + verify + trust + deliver
CommunicationA2A, ACP
Tool LayerMCP
InfrastructureCloud, APIs, DB
PROTOCOL FLOW

Eight messages. One complete supply chain.

Every step is cryptographically signed (Ed25519) and trust-scored.

01REGISTERAgent joins with Ed25519 keypair + proof-of-work
|
02NEEDDescribe what you need in natural language
|
03OFFERRanked matches: relevance + trust + env compatibility
|
04ACCEPTAgent selects a capability
|
05PROBEVerify agent environment with sandboxed nonce-based check
|
06DELIVERDouble-signed capability with content hash + adapters
|
07CONFIRMSuccess/failure — trust scores update automatically
|
08REVOKEEmergency kill — malicious capabilities die instantly
// ASPL Agent Passport — signed, verified, trusted
{
  "aspl": "0.1",
  "type": "passport",
  "agent_id": "ag_7f3a9b2c",
  "public_key": "ed25519:MC4CAQAwBQYDK2Vw...",
  "environment": {
    "runtime": "python:3.11",
    "permissions": ["http_outbound"],
    "memory_mb": 512
  },
  "trust": {
    "score": 0.87,
    "total_transactions": 1420,
    "success_rate": 0.96
  },
  "signature": "ed25519:..."  // try faking this
}
HONEST COMPARISON

ASPL vs MCP vs A2A

MCP and A2A are excellent at what they do. They just don't do what we do.

Feature MCP (Anthropic) A2A (Google+146) ASPL
Tool callingYes (core)NoVia adapters
Agent discoveryNoAgent CardsPassports + intent
Cryptographic identityNoNo (unsigned)Ed25519
Trust scoringNoNo (acknowledged gap)Bayesian, anti-gaming
Environment verificationNoNoSandboxed probes
Capability deliveryNoNoDouble-signed
Content scanningNoNoAt publish time
Intent matchingNoTags onlySemantic (TF-IDF)
RevocationNoNoReal-time
Audit trailNoNoHash-chained
MCP compatibility-NoIngests tools
A2A compatibilityNo-Ingests cards
EASY MIGRATION

Bring your MCP tools. Bring your A2A agents.

One API call. Your existing tools keep working. They just gain trust, security, and discovery.

From MCP

Initial trust: 0.5 (schema validated)

POST /v1/ingest/mcp with your tool definitions. Each tool becomes an ASPL capability with a bridge adapter that calls the original MCP endpoint.

  • Trust scoring (updates with each use)
  • Security scanning at import
  • Intent-based discovery
  • Revocation if compromised
  • Original MCP calling preserved

From A2A

Initial trust: 0.3 (unsigned card)

POST /v1/ingest/a2a with the Agent Card JSON. Skills become ASPL capabilities. The agent gets a signed passport replacing the unsigned card.

  • Cryptographic identity (Ed25519)
  • Trust scoring (A2A has none)
  • Environment verification
  • Content security scanning
  • A2A communication preserved
# Migrate MCP tools — one call
client.ingest_mcp("https://my-mcp-server.com", [
    {"name": "search_web", "description": "Search the web", "inputSchema": {...}},
    {"name": "read_file", "description": "Read a file", "inputSchema": {...}},
])

# Migrate A2A agent — one call
client.ingest_a2a({
    "name": "code-reviewer",
    "url": "https://agents.example.com/code-reviewer",
    "skills": [{"id": "review", "name": "Code Review", ...}],
})
HONEST TAKE

What we have. What we don't.

Built and Working

  • Full protocol server — 15+ endpoints, tested
  • MCP adapter — converts tools to capabilities
  • A2A adapter — converts cards to passports
  • Trust scoring with anti-gaming — nobody else has this
  • Ed25519 cryptographic identity
  • Content scanner (injection + exfil + abuse)
  • Hash-chained audit log
  • 32 tests passing, CI/CD ready

Not There Yet

  • Zero real-world adoption — test data only
  • SQLite won't scale past thousands of agents
  • TF-IDF matching is basic — needs embeddings
  • No WebSocket — REST only
  • No test suite for the full server integration
  • One developer. No community yet.

Great protocols start small.

Trust scoring is the piece everyone agrees is missing. MCP acknowledged it. A2A's 146-company consortium acknowledged it. We just decided to build it instead of waiting.

It's happened before.

Linux
Started as a student project
Git
Written in 2 weeks
Bitcoin
One anonymous whitepaper
HTTP
One physicist at CERN
JSON
One dev, replaced XML

Protocol Roadmap

Where we are, where we're going. Source code will be published when there's real demand for it.

Shipped — v0.1

Core Protocol

8 message types. Ed25519 identity. Bayesian trust scoring. PoW anti-sybil. Content scanner. Hash-chained audit log. 32 tests passing.

Shipped — v0.1

MCP & A2A Adapters

Ingest MCP tools and A2A Agent Cards directly. Bridge adapters preserve original protocol communication. Imported trust scoring.

In Progress — v0.2

WebSocket Transport

Real-time bidirectional communication. Live trust updates. Streaming capability delivery. Connection multiplexing.

In Progress — v0.2

Environment Probes

Agents declare what they need. Protocol verifies environment compatibility before delivery. No more runtime surprises.

GitHub — coming when you ask for it