AGA-mcp-server
The AGA-mcp-server is a cryptographic governance and compliance enforcement platform for autonomous AI agents, implementing the Attested Governance Artifact (AGA) protocol. It acts as a zero-trust policy enforcement point, providing tamper-evident attestation, behavioral monitoring, and signed auditing.
Policy Enforcement & Attestation
Intercepts MCP client tool calls, evaluates them against sealed policies (allowlist, denylist, audit_only; permissive/standard/restrictive profiles), and produces Ed25519-signed receipts
Creates sealed Policy Artifacts, measures subject state against sealed references, verifies artifact signatures, and supports mid-session revocation with TERMINATED or SAFE_STATE transitions
Behavioral Monitoring & Drift Detection
Define baselines (permitted tools, rate limits, forbidden sequences) and detect behavioral drift via continuous or on-demand measurement
Tamper-Evident Auditing & Chain Management
Bootstrap a continuity chain with a genesis event, log all events (behavioral, delegations, receipts, revocations, attestations), and retrieve/verify chain history
Evidence Bundles & Offline Verification
Export verifiable evidence bundles (artifact + receipts + Merkle proofs) and perform 4-step offline bundle verification
Identity, Access & Key Management
Manage Ed25519 keys with pinned issuer keys, TTL re-attestation, and auditable key rotation
Set verification tiers (BRONZE, SILVER, GOLD) and delegate constrained policies to sub-agents (scope only diminishes, never escalates)
Privacy & Disclosure
Privacy-preserving claim disclosure with auto-substitution (PROOF_ONLY, REVEAL_MIN, REVEAL_FULL modes)
Developer & Diagnostics Tools
Full lifecycle demo tool (attest → measure → drift → revoke → bundle → verify) for scenarios like drone or SCADA
Server info, portal state inspection, and quarantine/forensic capture status
3 resources (protocol spec, sample bundle, crypto primitives) and 3 prompts for demos/reports
The server uses a two-process boundary where the proxy holds signing keys (the client holds none), ensuring fail-closed enforcement and covering all 12 CoSAI MCP security threat categories.
@attested-intelligence/aga-mcp-server v2.1.0
MCP server and governance proxy implementing the Attested Governance Artifact (AGA) protocol - cryptographic compliance enforcement for autonomous AI systems.
What It Does
This server acts as a Portal (zero-trust Policy Enforcement Point) for AI agents. Every tool call is attested, measured against a sealed cryptographic reference, and logged to a tamper-evident continuity chain with signed receipts.
20 tools, 3 resources, 3 prompts, governance proxy, 199 tests
Governance Proxy (New in v2.1.0)
Sits between any MCP client (OpenClaw, Claude Desktop, etc.) and any downstream MCP server. Intercepts every tools/call, evaluates it against a sealed policy, and produces Ed25519-signed receipts in the canonical Ed25519-SHA256-JCS format (compatible with the Python SDK, gateway, and browser verifier).
MCP Client --> AGA Proxy (:18800) --> Downstream MCP Server
|
+-- Policy evaluation
+-- Signed receipt per tool call
+-- Merkle tree + evidence bundleProxy Quick Start
# Start with a downstream MCP server
npx tsx src/proxy/index.ts start --upstream "node server.js" --profile standard
# Policy profiles: permissive, standard, restrictive
npx tsx src/proxy/index.ts start --upstream-url http://localhost:3000 --profile restrictive
# Export evidence bundle (verifiable at attestedintelligence.com/verify)
npx tsx src/proxy/index.ts export --output bundle.json
# Verify a bundle
npx tsx src/proxy/index.ts verify bundle.jsonProxy Features
Policy modes: allowlist, denylist, audit_only
Rate limiting: per-tool calls/minute with sliding window
Path constraints: restrict file tools to allowed prefixes
Denied patterns: block dangerous argument patterns
Receipt format: Ed25519-SHA256-JCS (canonical across all AGA SDKs)
Evidence bundles: verifiable at
attestedintelligence.com/verifyTwo-process boundary: proxy holds all signing keys, client holds none
20 MCP Tools
# | Tool | Description |
1 |
| Server identity, keys, portal state, framework alignment |
2 |
| Initialize continuity chain with genesis event |
3 |
| Attest subject, generate sealed Policy Artifact |
4 |
| Measure subject, compare to sealed ref, generate receipt |
5 |
| Verify artifact signature against issuer key |
6 |
| Start/restart behavioral monitoring with baseline |
7 |
| Current portal enforcement state and TTL |
8 |
| Trigger measurement with specific type |
9 |
| Generate signed measurement receipt manually |
10 |
| Package artifact + receipts + Merkle proofs |
11 |
| 4-step offline bundle verification |
12 |
| Privacy-preserving disclosure with auto-substitution |
13 |
| Get chain events with optional integrity verification |
14 |
| Quarantine state and forensic capture status |
15 |
| Mid-session artifact revocation |
16 |
| Set verification tier (BRONZE/SILVER/GOLD) |
17 |
| Full lifecycle: attest, measure, checkpoint, verify |
18 |
| Behavioral drift detection (tool patterns) |
19 |
| Constrained sub-agent delegation (scope only diminishes) |
20 |
| Key rotation with chain event |
3 Resources
Resource | URI | Description |
Protocol Spec |
| Full protocol specification with SPIFFE alignment |
Sample Bundle |
| Sample evidence bundle documentation |
Crypto Primitives |
| Cryptographic primitives documentation |
3 Prompts
Prompt | Description |
| 4-phase NCCoE lab demo with behavioral drift |
| Session governance summary report |
| Drift event analysis and remediation |
CoSAI MCP Security Threat Coverage
The AGA MCP Server addresses all 12 threat categories identified in the CoSAI MCP Security whitepaper (Coalition for Secure AI / OASIS, January 2026).
CoSAI Category | Threat Domain | AGA Governance Mechanism |
T1: Improper Authentication | Identity & Access | Ed25519 artifact signatures, pinned issuer keys, TTL re-attestation, key rotation chain events |
T2: Missing Access Control | Identity & Access | Portal as mandatory enforcement boundary, sealed constraints, delegation with scope diminishment |
T3: Input Validation Failures | Input Handling | Runtime measurement against sealed reference, behavioral drift detection |
T4: Data/Control Boundary Failures | Input Handling | Behavioral baseline (permitted tools, forbidden sequences, rate limits), phantom execution forensics |
T5: Inadequate Data Protection | Data & Code | Salted commitments, privacy-preserving disclosure with substitution, inference risk prevention |
T6: Missing Integrity Controls | Data & Code | Content-addressable hash binding, 10 measurement embodiments, continuous runtime verification |
T7: Session/Transport Security | Network & Transport | TTL-based artifact expiration, fail-closed on expiry, mid-session revocation, Ed25519 signed receipts |
T8: Network Isolation Failures | Network & Transport | Two-process architecture, agent holds no credentials, NETWORK_ISOLATE enforcement action |
T9: Trust Boundary Failures | Trust & Design | Enforcement pre-committed by human authorities in sealed artifact, not delegated to LLM |
T10: Resource Management | Trust & Design | Per-tool rate limits in behavioral baseline, configurable measurement cadence (10ms to 3600s) |
T11: Supply Chain Failures | Operational | Content-addressable hashing at attestation, runtime hash comparison blocks modified components |
T12: Insufficient Observability | Operational | Signed receipts, tamper-evident continuity chain, Merkle anchoring, offline evidence bundles |
Full mapping details available via the aga://specification resource.
Quick Start
npm install && npm run build && npm testConnect to an MCP Client
Add to your MCP client config:
{
"mcpServers": {
"aga": { "command": "node", "args": ["/path/to/aga-mcp-server/dist/index.js"] }
}
}Architecture
MCP Client
| JSON-RPC over stdio
v
src/server.ts - 20 tools + 3 resources + 3 prompts
|
+-- src/tools/ 20 individual tool handlers
+-- src/core/ Protocol logic (artifact, chain, portal, etc.)
+-- src/crypto/ Ed25519 + SHA-256 + Merkle + canonical JSON
+-- src/middleware/ Zero-trust governance PEP
+-- src/storage/ In-memory + optional SQLite
+-- src/resources/ Protocol docs + crypto primitives
+-- src/prompts/ Demo + report + analysis prompts
+-- src/proxy/ Governance proxy (NEW in v2.1.0)
+-- src/adapters/ OpenClaw config adapterTest Coverage
Suite | Tests | What |
Crypto | 33 | SHA-256, Ed25519, Merkle, salt, canonical, keys |
Core | 56 | Artifact, chain, portal, governance, behavioral, delegation, privacy, revocation, fail-closed |
Tools | 25 | All 20 tool handlers |
Integration | 38 | Bundle tamper, lifecycle, performance, NCCoE demo, crucible compatibility |
Proxy | 40 | Policy evaluator, round-trip, cross-verification, OpenClaw adapter |
Total | 199 |
License
MIT - Attested Intelligence Holdings LLC
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/attestedintelligence/aga-mcp-server'
If you have feedback or need assistance with the MCP directory API, please join our Discord server