Agent Identity Bridge (AIB)

One identity. Every protocol. Full audit trail.

AIB is an open-source protocol that gives AI agents a single portable identity across MCP (Anthropic), A2A (Google), ANP, and AG-UI — with W3C DID v1.1, Verifiable Credentials, and EU AI Act compliance built in.

The problem

Each AI protocol invented its own identity system. An agent operating across MCP + A2A + AG-UI has three separate identities with zero link between them. Cross-protocol auditing is impossible, credential management is painful, and EU AI Act compliance is a nightmare.

Quick start

pip install agent-identity-bridge
python -c "from aib import PassportService; print('AIB ready')"

Run the interactive demo (15 seconds, zero config)

curl -sO https://raw.githubusercontent.com/tntech-consulting/agent-identity-bridge/main/examples/quickstart.py
python quickstart.py

AIB Cloud (managed SaaS)

from aib.cloud import AIBCloud

client = AIBCloud("aib_sk_live_...")

passport = client.create_passport("my-bot", protocols=["mcp", "a2a", "ag_ui"])
mcp_card = client.translate(a2a_card, "a2a_agent_card", "mcp_server_card")
client.create_policy("deliverable_gate", {"required_capabilities": ["tests_passed"]})
client.create_webhook("https://your-app.com/hooks", events=["passport.created"])

Local protocol (self-hosted)

from aib.passport import PassportService, McpBinding, A2aBinding

svc = PassportService(secret_key="your-secret")
passport, token = svc.create_passport(
    org_slug="mycompany", agent_slug="booking-agent",
    capabilities=["booking", "scheduling"],
    bindings={"mcp": McpBinding(auth_method="oauth2"), "a2a": A2aBinding(auth_method="bearer")},
)

Framework integrations

from aib.integrations import get_langchain_tools   # LangChain
from aib.integrations import get_crewai_tools       # CrewAI
from aib.integrations import get_openai_agents_tools # OpenAI Agents SDK

Key features

• Portable identity: One passport, valid on MCP, A2A, ANP, AG-UI
• Credential translation: 6 paths across 4 formats (< 1ms)
• W3C DID v1.1: did:web + did:key, resolvable by any Universal Resolver
• W3C Verifiable Credentials: Ed25519Signature2020 proof, StatusList2021 revocation
• EU AI Act compliance: 7 structured fields (intent, risk_level, human_oversight, etc.) in every signed receipt
• Policy engine: 12 rule types — deliverable gates, separation of duties, capability enforcement
• Ed25519 audit trail: Signed receipts with SHA-256 hash chaining, AES-256 encrypted keys
• OIDC federation: Google, Microsoft Entra, Okta, Auth0 — bring your own IdP
• Webhooks: 4 event types, HMAC-SHA256 signed payloads
• Rate limiting: 30 req/min per IP (HTTP) + policy-based per org
• Intent inference: LLM-powered analysis of agent behavior patterns
• Delegation chains: Parent → child with scope narrowing enforcement
• Framework integrations: LangChain, CrewAI, OpenAI Agents SDK

AIB Cloud — Managed SaaS

19 Edge Functions, 13 API endpoints, 17 tables with RLS.

Dashboard: aib-tech.fr/dashboard EU AI Act Compliance Kit: aib-tech.fr/compliance

How AIB relates to existing solutions

AIB is a standard, not a product. It bridges protocols, it doesn't compete with IAM vendors.

• MCP connects agents to tools. AIB connects agents to all protocols.
• A2A coordinates agents. AIB gives each agent an identity usable everywhere.
• Okta/Entra authenticates humans. AIB bridges their tokens to agent passports via OIDC federation.
• SailPoint discovers and governs agents. AIB gives them portable identities across protocols.

No European competitor exists for cross-protocol agent identity + DID + VCs + EU AI Act compliance. See aib-tech.fr/why for full positioning.

Contributing

Apache 2.0 licensed. Contributions welcome.

git clone https://github.com/tntech-consulting/agent-identity-bridge.git
cd agent-identity-bridge
pip install -e ".[dev]"
pytest  # 1,054 tests

Author

Thomas Nirennold — TNTECH CONSULTING SAS (SIREN 993811157), Paris, France

Protocol · Why AIB · Compliance Kit · Blog · PyPI

License

Apache 2.0 — see LICENSE