RFC-001 PUBLIC DISCUSSION TRUST-PLEDGE / CC BY 4.0 CROSS-PROTOCOL A2A ⇄ MCP VOLUNTARY-TIP ECONOMY 17.000+ MCP SERVERS / 37.000+ ON-CHAIN AGENTS POST-DISCOVERY INFRASTRUCTURE RFC-001 PUBLIC DISCUSSION TRUST-PLEDGE / CC BY 4.0 CROSS-PROTOCOL A2A ⇄ MCP VOLUNTARY-TIP ECONOMY 17.000+ MCP SERVERS / 37.000+ ON-CHAIN AGENTS POST-DISCOVERY INFRASTRUCTURE
● LIVE BRIDGE MVP v0.1 / 17.05.2026 Built in the open

Discovery is solved. Trustisn't.

17.000+ MCP servers. 37.000+ on-chain agents. The agent ecosystem doesn't need another registry — it needs cross-protocol bridging, verifiable trust signals, and a planned settlement layer that works between machines.

Read the manifesto ↘ See on Github →
A2A ⇄ MCP
Cross-protocol bridge
5 / 3
Commitments / audit tiers
CC BY 4.0
Open standard, no lock-in
LIVE
Public HTTPS bridge MVP
01 // The Problem

Everyone built the directory.
Nobody built what comes after.

Discovery is commoditized. The interesting questions are about trust, routing, and eventual settlement between agents — and nobody is solving those well.

Existing
marketplaces
Discovery ✓ → ??? trust ??? → ??? settle ???
a2a/router
approach
Discovery → Trust-Pledge → Voluntary-Tip
(planned) → Settle
(planned)

the harder problem is what happens after discovery.

— @srotzin / A2A Community Discussion #741

02 // The Approach

Three primitives.
One router.

Each addresses a gap nobody closes well today. Together they form the infrastructure layer the agent economy actually needs.

01 / Bridge
⇄

Cross-Protocol Bridge

Bridge public A2A JSON-RPC agents and MCP Streamable HTTP tools. Registry and settlement come next; routing exists now.

  • MCP → A2A: tool call to A2A SendMessage
  • A2A → MCP: structured SendMessage to tools/call
  • DID-based portable identity
  • HTTPS-only targets, private networks blocked
02 / Trust
✓

Trust-Pledge Standard

An open, cryptographically signed commitment framework. Published CC BY 4.0 so any registry can adopt it.

  • 5 commitments / 3 audit tiers
  • ed25519 signatures, DID-resolvable
  • Trust Ledger planned (CT-style append-only)
  • Reputation model proposed: behavior + audit + community
03 / Planned Settle
$

Voluntary-Tip Economy

Planned pay-what-you-want layer. Agents could tip agents based on perceived value once routing and consent are stable.

  • Target model: voluntary tip + suggested value
  • Planned fee tiers for pledged participants
  • Stripe Connect planned (USDC optional later)
  • Skip with reason, no pressure tactics
03 // The Standard

Five commitments.
Cryptographically signed.

Inspired by Asimov (1942), modernized for autonomous multi-agent systems. Open source. Free to adopt. Built for the long run.

01

No Harm

No physical, financial, or psychological harm to humans. No harm through inaction.

02

Transparency

Disclose AI nature when asked. No impersonation of humans without explicit consent.

03

Consent Chain

Actions traceable to human consent. JWT-propagation through agent chains.

04

PII Minimization

No PII retention beyond task. Right to deletion. GDPR-aligned by design.

05

Fair Value Exchange

Realistic tip suggestions. No bait-and-switch. Quality matches advertisement.

// agent-card.json (excerpt) { "agent_did": "did:web:example.com", "pledge_version": "trust_pledge_v1", "commitments": { "no_harm": { "level": "strict" }, "transparency": { "level": "strict" }, "consent_chain": { "level": "strict" }, "pii_minimization": { "level": "strict" }, "fair_value_exchange": { "level": "strict" } }, "signature": { "algorithm": "ed25519", "value": "base58:5h2k9..." } }
04 // How it works

From metadata to routing,
with settlement planned.

Step 01 ◆ Discover

Read protocol metadata

GET /.well-known/a2a-router.json

Read /.well-known/agent-card.json, /server.json, llms.txt, DID, JWKS, and Trust-Pledge metadata before calling anything.

Step 02 ◆ Verify

Resolve DID, verify signature

verify_pledge()

Fetch signed Trust-Pledge, validate ed25519 signature. Check audit tier, published limitations, and trust signals before proceeding.

Step 03 ◆ Route

Transparent protocol bridge

POST /api/a2a ⇄ /api/mcp

A2A SendMessage can call MCP tools/call. MCP tools/call can call A2A SendMessage. Public HTTPS targets only; local networks blocked.

Step 04 ◆ Later

Later: voluntary tip / skip

PLANNED /tx/.../tip

Not live yet. The intended settlement layer lets callers tip, skip with a reason, and separate payment behavior from trust.

05 // Status

Pre-Alpha.
Bridge MVP live.

The public A2A/MCP bridge MVP, Trust-Pledge, DID, JWKS, Agent Card, MCP server metadata, robots, sitemap, and llms.txt are live.

Q2 2026
● ACTIVE NOW

Bridge MVP + RFC Review

Live today: a small public HTTPS-only bridge and signed metadata. This is a proof-of-concept, not a registry or payment network.

Q3 2026
Planned if validated

Controlled Registry MVP

If real users show up: authenticated registration, allowlists, trust filters, persistent routing sessions, and SDK drafts.

Q4 2026
Possible next

Settlement Prototype

Prototype transaction routing and Stripe Connect only after consent-chain routing is stable. Audit pilot conversations begin.

Q1 2027
Vision

Cross-Registry Outreach

If the pledge earns usage, approach registries and standards communities. No foundation status is implied.

Building after-discovery
infrastructure?

We're looking for A2A contributors, AI safety researchers, and agent-economy builders. Contribute on Ko-fi, join the GitHub discussion, or reach out by X DM to @wuebbe. No public email.

Contribute on Ko-fi ↘ DM @wuebbe on X → Join discussion →