Kryptografischer Audit-Trail für KI-Agenten via Model Context Protocol
Related MCP server: DCL Evaluator
Der PiQrypt MCP-Server bietet Zugriff via Model Context Protocol auf PiQrypt — den post-quantenkryptografischen Audit-Trail für KI-Agenten.
Anwendungsfälle:
🤖 KI-Agenten: Jede Entscheidung mit einem kryptografischen Nachweis signieren
📊 n8n-Workflows: Audit-Trail zu Automatisierungs-Workflows hinzufügen
🏦 Trading-Bots: SEC/FINRA-Compliance für automatisierten Handel
👥 HR-Automatisierung: DSGVO-konforme KI-Einstellungsentscheidungen
🏥 KI im Gesundheitswesen: HIPAA-Audit-Trail für medizinische Entscheidungen
1. PiQrypt installieren (erforderlich — Python 3.8+)
pip install piqryptDer MCP-Server delegiert alle kryptografischen Operationen an das Python-Paket piqrypt.
Wenn es nicht installiert ist, gibt der Server bei jedem Tool-Aufruf einen klaren Fehler zurück.
2. Den MCP-Server installieren (Node.js 18+)
npm install -g @piqrypt/mcp-servernpx @piqrypt/mcp-servergit clone https://github.com/piqrypt/piqrypt-mcp-server
cd piqrypt-mcp-server
npm install
npm run buildStandardmäßig verwendet der Server python3 (Linux/Mac) oder python (Windows).
Wenn piqrypt in einer virtuellen Umgebung installiert ist, setzen Sie diese Variable, um auf den richtigen Interpreter zu verweisen:
Windows
set PIQRYPT_PYTHON=C:\path\to\venv\Scripts\python.exeLinux / Mac
export PIQRYPT_PYTHON=/path/to/venv/bin/pythonUm dies dauerhaft zu machen, fügen Sie es Ihrer MCP-Client-Konfiguration hinzu:
{
"mcpServers": {
"piqrypt": {
"command": "piqrypt-mcp-server",
"args": [],
"env": {
"PIQRYPT_PYTHON": "/path/to/venv/bin/python"
}
}
}
}Hinzufügen zu ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"piqrypt": {
"command": "piqrypt-mcp-server",
"args": []
}
}
}n8n MCP-Integration installieren
PiQrypt MCP-Server zur Konfiguration hinzufügen
In Workflows über den MCP-Node verwenden
Client | Version | Hinweise |
Jeder MCP-kompatible Client | MCP-Spezifikation 2024-11+ | stdio-Transport |
n8n | 1.88+ | via MCP-Node |
Cursor | beliebig | zu MCP-Einstellungen hinzufügen |
VS Code | beliebig | zu MCP-Einstellungen hinzufügen |
Continue | beliebig | zu MCP-Einstellungen hinzufügen |
Windsurf | beliebig | zu MCP-Einstellungen hinzufügen |
Plattform | Integration | Anwendungsfall |
n8n | MCP-Node (nativ) | No-Code Audit-Trail |
Make.com | HTTP-Modul | Webhook-getriggertes Stempeln |
Zapier | Webhooks + HTTP | Einfache Ereignisaufzeichnung |
Jeder Tool-Aufruf durchläuft dieselben 4 Operationen — Stempeln, Verifizieren, Exportieren, Suchen. Hier ist, was das in der Praxis je nach Kontext bedeutet:
Automatisierter Handel / Finanzen Jeder Agent, der Aufträge erteilt, Portfolios umschichtet oder Transaktionen auslöst, kann jede Entscheidung vor der Ausführung stempeln. Die signierte Kette ist für SEC/FINRA- Audits ohne zusätzliche Infrastruktur exportierbar.
HR- und Einstellungsautomatisierung Jeder Workflow, der Kandidaten bewertet, Lebensläufe punktet oder Bewerber weiterleitet, kann jede Entscheidung stempeln. Bietet einen DSGVO Art. 22-konformen Audit-Trail für KI-gestützte Einstellungsprozesse — wer hat was wann entschieden und welche Daten wurden verwendet (gehasht, niemals im Klartext gespeichert).
Content- und Publishing-Pipelines Jeder Agent, der Inhalte entwirft, genehmigt oder veröffentlicht, kann jeden Schritt stempeln. Nützlich, wenn mehrere KI-Agenten zusammenarbeiten und Sie die Zuordnung nachweisen müssen — welcher Agent hat was in welcher Reihenfolge geschrieben.
DevOps und CI/CD Jeder Agent, der Deployments auslöst, Branches zusammenführt oder Geheimnisse rotiert, kann jede Aktion stempeln. Bietet eine fälschungssichere Aufzeichnung von Infrastrukturänderungen, die von autonomen Agenten vorgenommen wurden.
Gesundheitswesen und medizinische KI Jeder Diagnose- oder Triage-Agent kann jede Empfehlung stempeln. Bietet einen HIPAA-konformen Audit-Trail, der jede KI-Ausgabe mit einer verifizierbaren Agentenidentität verknüpft.
Das gemeinsame Muster in allen Fällen:
[Agent makes decision]
↓
piqrypt_stamp_event ← sign + chain
↓
[Agent executes action]
↓
piqrypt_export_audit ← portable proof, verifiable
without PiQrypt installedpiqrypt_stamp_eventSignieren Sie eine KI-Entscheidung mit einem kryptografischen Nachweis.
Parameter:
agent_id (Zeichenkette, erforderlich): Agenten-Identifikator
payload (Objekt, erforderlich): Entscheidungsdaten
previous_hash (Zeichenkette, optional): Hash des vorherigen Ereignisses zur Verkettung
Beispiel:
const event = await mcp.call('piqrypt_stamp_event', {
agent_id: 'trading_bot_v1',
payload: {
action: 'buy',
symbol: 'AAPL',
quantity: 100,
price: 150.25
}
});Rückgabe:
{
"version": "AISS-1.0",
"agent_id": "trading_bot_v1",
"timestamp": 1739382400,
"nonce": "uuid-...",
"payload": { ... },
"previous_hash": "sha256:...",
"signature": "base64:..."
}piqrypt_verify_chainÜberprüfen Sie die Integrität der Ereigniskette.
Parameter:
events (Array, erforderlich): Zu überprüfende Ereignisse
Beispiel:
const result = await mcp.call('piqrypt_verify_chain', {
events: [event1, event2, event3]
});Rückgabe:
{
"valid": true,
"events_count": 3,
"chain_hash": "sha256:...",
"errors": []
}piqrypt_export_auditExportieren Sie den Audit-Trail für Compliance-Zwecke.
Parameter:
agent_id (Zeichenkette, erforderlich): Zu exportierender Agent
certified (Boolesch): PiQrypt-Zertifizierung anfordern
output_format (Zeichenkette): json oder pqz
Beispiel:
const audit = await mcp.call('piqrypt_export_audit', {
agent_id: 'trading_bot_v1',
certified: true,
output_format: 'json'
});piqrypt_search_eventsSchnelle Suche über SQLite-Index.
Parameter:
event_type (Zeichenkette, optional): Nach Typ filtern
from_timestamp (Zahl, optional): Startzeit
to_timestamp (Zahl, optional): Endzeit
limit (Zahl): Maximale Ergebnisse (Standard: 100)
Beispiel:
const trades = await mcp.call('piqrypt_search_events', {
event_type: 'trade_executed',
from_timestamp: 1739300000,
limit: 50
});Jedes gestempelte Ereignis ist in Vigil sichtbar — PiQrypts lokales Monitoring-Dashboard.
Hinweis: Vigil wird nicht automatisch vom MCP-Server gestartet. Sie müssen es separat starten, bevor Sie das Dashboard öffnen.
piqrypt vigil
# → http://localhost:8421Der kostenlose Tarif beinhaltet: Kettenintegrität, VRS-Risiko-Score, 7-Tage-Verlauf, KRITISCHE Warnungen. Upgrade auf Pro für 90-Tage- Verlauf, TrustGate-Governance und post-quanten Signaturen.
Agenten werden beim ersten Stempeln automatisch erstellt. Um Agenten anzuzeigen und zu löschen:
Vigil starten: piqrypt vigil
http://localhost:8421 öffnen
Zur Ansicht All Agents gehen
Die zu löschenden Agenten auswählen → auf ✕ Delete selected klicken
Bestätigen — Vigil kehrt zum Willkommensbildschirm zurück, wenn keine Agenten mehr vorhanden sind
Agenten werden auf Ihrem Computer unter
~/.piqrypt/agents/gespeichert. Das Löschen eines Agenten entfernt dauerhaft seine Schlüssel und seinen Ereignisverlauf.
┌─────────────────────────────────────┐
│ MCP Client (any MCP-compatible client) │
│ ↓ JSON-RPC over stdio │
├─────────────────────────────────────┤
│ MCP Server (TypeScript/Node.js) │ ← No crypto here
│ ↓ subprocess call │
├─────────────────────────────────────┤
│ Python Bridge (bridge.py) │
│ ↓ invokes CLI │
├─────────────────────────────────────┤
│ PiQrypt CLI (Python) │
│ ↓ uses │
├─────────────────────────────────────┤
│ Core Crypto (aiss package) │ ← All crypto here
│ • Ed25519 / Dilithium3 │
│ • RFC 8785 canonical JSON │
│ • Hash chains │
└─────────────────────────────────────┘✅ Private Schlüssel niemals gegenüber der MCP-Ebene offengelegt ✅ Alle Kryptografie in Python (Ed25519, Dilithium3) ✅ Gleiche Sicherheit wie CLI (Prozessisolierung) ✅ RFC AISS-1.1 konform (identische Ausgabe) ✅ Eingabevalidierung vor dem Subprozess-Aufruf
[Webhook: price alert]
↓
[AI Decision: buy/sell?]
↓
[PiQrypt MCP: stamp decision] ← Audit trail
↓
[Execute trade API]
↓
[Database: store proof][Upload CV]
↓
[AI Agent: evaluate candidate]
↓
[PiQrypt MCP: stamp evaluation] ← GDPR compliance
↓
[Email HR team]# Build
npm run build
# Test bridge
python3 src/python/bridge.py stamp '{"agent_id":"test","payload":{"action":"test"}}'
# Test MCP server (manual)
node dist/index.js
# Then send MCP request via stdinError: piqrypt is not installed in this Python environmentDer vom MCP-Server verwendete Python-Interpreter kann das piqrypt-Paket nicht finden.
Lösung:
pip install piqryptWenn piqrypt in einer virtuellen Umgebung und nicht im System-Python installiert ist, setzen Sie PIQRYPT_PYTHON auf den korrekten Interpreter:
# Linux / Mac
export PIQRYPT_PYTHON=/path/to/venv/bin/python
# Windows
set PIQRYPT_PYTHON=C:\path\to\venv\Scripts\python.exeUm zu überprüfen, welches Python der Server verwenden wird:
# Linux / Mac
$PIQRYPT_PYTHON -c "import aiss; print('ok')"
# Windows
%PIQRYPT_PYTHON% -c "import aiss; print('ok')"Wir freuen uns über Beiträge! Siehe CONTRIBUTING.md.
MCP-Server → MIT-Lizenz - siehe LICENSE PiQrypt Core → kostenloser Tarif + kommerzielle Tarife
PiQrypt Core: https://github.com/piqrypt/piqrypt
MCP-Protokoll: https://modelcontextprotocol.io
n8n: https://n8n.io
Dokumentation: https://docs.piqrypt.com
Mit ❤️ gebaut von PiQrypt Inc.
Export the complete agent audit trail to a portable JSON archive. Set certified=true to request a PiQrypt CA signature for legal admissibility (eIDAS Art.26). The export is self-contained and verifiable without PiQrypt installed.
| Name | Required | Description | Default |
|---|---|---|---|
| agent_id | Yes | Agent ID to export | |
| certified | No | Create certified export (requires Pro license) | |
| output_format | No | Output format (json or encrypted pqz archive) | json |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Discloses that export is self-contained and verifiable, and mentions Pro license requirement. Does not detail side effects or reading behavior; with no annotations, this is moderate.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences, each adding essential information: purpose and certification detail. No unnecessary words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Lacks output schema but describes the result as a portable JSON archive. Missing details on error conditions or what the audit trail contains, but adequate for a simple export.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so baseline is 3. Description adds value for the certified parameter (legal admissibility) and clarifies the export scope (complete trail).
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool exports a complete agent audit trail to JSON, using a specific verb and resource. It distinguishes well from sibling tools like piqrypt_search_events.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Provides context on when to use certified=true and notes the export is self-contained. Lacks explicit when-not-to-use or alternatives, but siblings are distinct actions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Search the agent's cryptographic event history by type, time range, or session. Returns signed events with chain metadata. Use to reconstruct what an agent did during a specific period.
| Name | Required | Description | Default |
|---|---|---|---|
| event_type | No | Filter by event type (e.g., "trade_executed", "decision_made") | |
| from_timestamp | No | Start timestamp (Unix UTC seconds) | |
| to_timestamp | No | End timestamp (Unix UTC seconds) | |
| limit | No | Maximum number of results |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided; description states it returns signed events with chain metadata, implying read-only. However, it does not mention authentication, rate limits, pagination behavior, or what happens with empty results.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Three concise sentences with no filler. The action verb is front-loaded, and every sentence adds useful context.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
No output schema; description gives a brief idea of return values (signed events with chain metadata) but omits details like event structure or metadata fields. The mention of session without a corresponding parameter reduces completeness.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so the description adds minimal value beyond repeating filter options. It mentions session but the schema lacks a session parameter, creating slight ambiguity.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Clearly states verb (search), resource (cryptographic event history), and scope (by type, time range, or session). Distinguishes from siblings like piqrypt_stamp_event (create) and piqrypt_export_audit (export).
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Provides a clear use case ('reconstruct what an agent did during a specific period'), but does not explicitly state when not to use or mention alternatives among siblings.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Create a tamper-proof cryptographic record of an agent decision. Signs the event with Ed25519, links it to the previous event in a hash chain (AISS v2.0). Call this after every significant agent action. Required for GDPR Art.22, EU AI Act Art.13, HIPAA audit trail, SEC/FINRA trading compliance.
| Name | Required | Description | Default |
|---|---|---|---|
| agent_id | Yes | Agent identifier (e.g., "trading_bot_v1", "hr_decision_engine") | |
| payload | Yes | Event payload containing decision data (JSON object) | |
| previous_hash | No | Optional hash of previous event for chain integrity |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Without annotations, the description discloses key behaviors: tamper-proof record creation, Ed25519 signing, and hash chain linking. It lacks details on side effects, idempotency, or error handling, but covers essential traits.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is concise at two sentences, front-loading the purpose and including key details. It is efficient but could be slightly more structured by grouping regulations separately.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a tool with 3 parameters and no output schema, the description covers purpose, usage, and behavioral details adequately. The lack of return value description is a minor shortfall, but overall it is fairly complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema already describes all three parameters with 100% coverage. The description adds no additional semantics beyond the schema, so a baseline score of 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool creates a cryptographic record, specifying the signing algorithm (Ed25519) and chaining mechanism (AISS v2.0), which distinguishes it from sibling tools for export, search, and verification.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explicitly advises 'Call this after every significant agent action' and lists regulatory requirements, providing strong contextual cues. However, it does not explicitly exclude cases where the tool should not be called.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Verify that an agent's decision history is intact and untampered. Detects modified events, missing events, hash chain breaks, and forks. Call before trusting any historical agent output.
| Name | Required | Description | Default |
|---|---|---|---|
| events | Yes | Array of PiQrypt events to verify |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description adequately describes what the tool detects (tampering evidence). However, it lacks details on output/return behavior (e.g., error vs result) and side effects, which would improve transparency.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences, no wasted words, front-loaded with the main action and key details.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple tool with full schema coverage, the description covers purpose and usage. It lacks specification of return value (e.g., boolean, report), but overall is complete enough for an agent to invoke correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% and the description does not add additional meaning beyond the schema's 'Array of PiQrypt events to verify'. Baseline score of 3 is appropriate as the schema already carries the parameter semantics.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's purpose: verifying integrity of an agent's decision history. It lists specific detection capabilities (modified events, missing events, etc.) and distinguishes itself from siblings by its verification role.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description includes explicit usage guidance: 'Call before trusting any historical agent output.' While it doesn't explicitly exclude scenarios or name alternatives, the clear context is sufficient for an agent to decide when to invoke this tool.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v1.5.3piqrypt_export_auditpiqrypt_search_eventspiqrypt_stamp_eventpiqrypt_verify_chainScored across 4 tools
Each tool maps to a distinct action in the audit workflow: stamping, exporting, searching, and verifying. The descriptions make the boundaries clear, and even the slight overlap between search and export is resolved by their different purposes.
Every tool follows the same snake_case convention with a shared 'piqrypt_' prefix and a clear verb_object pattern, such as stamp_event and verify_chain. The small singular/plural variation is natural and does not hurt predictability.
Four tools is well-scoped for a focused cryptographic audit-log server. Each tool covers a necessary part of the lifecycle and none feels redundant or missing.
The tool surface covers the full lifecycle of an append-only audit trail: create, query, export, and verify. Missing update/delete operations are appropriate for a tamper-proof audit system, so there are no obvious dead ends.
Bitcoin-anchored, tamper-evident audit log for AI agents — record, disclose and verify actions.
Tamper-evident proof creation and verification for AI agents via MCP, A2A, and REST.
Register every AI agent, log every action, prove it. EU AI Act compliance built in.
Bitcoin-anchored, tamper-evident audit-permanence layer for AI agents, FRE 902(13)/(14)-shaped.