Gives Claude six tools to write, search, and verify a local SHA-256 hash chain for agent events. You can log 16 event types (payments, tasks, disputes, identity checks, messages), retrieve full histories per agent, search by severity or date range, and run integrity checks that show exactly which link broke. Every event cryptographically includes the previous hash, so tampering is detectable. Data lives in a local JSON file at ~/.agentaudit/chain.json with atomic writes to prevent corruption. Reach for this when you need an immutable audit trail for multi-agent systems or compliance logging without sending data off-machine. Requires a $19/month subscription.
Your agent now has an immutable, tamper-evident log of every action it takes — each event cryptographically linked to the previous one so any modification breaks the chain and is immediately detectable.
info, notice, warning, error, critical — for compliance triageRequires: Python 3.10+, mcp and fastmcp packages.
pip install mcp fastmcp
Claude Desktop — add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"agent-audit": {
"command": "python",
"args": ["/absolute/path/to/agent-audit-mcp/server.py"]
}
}
}
Cursor — add to .cursor/mcp.json in your project root:
{
"mcpServers": {
"agent-audit": {
"command": "python",
"args": ["/absolute/path/to/agent-audit-mcp/server.py"]
}
}
}
Data is stored locally at ~/.agentaudit/chain.json. No data leaves your machine.
| Tool | Description | Key params |
|---|---|---|
audit_log | Log an event and append it to the hash chain | event_type (one of 16), agent_id, severity, details_json |
audit_get_event | Retrieve full event details by ID | event_id |
audit_search | Filter events by agent, type, severity, or date range | agent_id, event_type, severity, from_date, to_date, max_results |
audit_get_agent_history | All events for one agent, newest first | agent_id, max_events |
audit_verify_chain | Check SHA-256 chain integrity between two events; returns per-event PASS/FAIL | from_event_id, to_event_id |
audit_stats | Summary: total events, breakdown by type/severity, top 5 active agents | — |
Supported event types: task_created, task_assigned, task_completed, task_failed, payment_sent, payment_received, payment_refunded, dispute_raised, dispute_resolved, identity_verified, capability_used, agent_registered, agent_rated, agent_offline, message_sent, message_received
Audit data is written atomically (write-to-temp, then os.replace) to prevent partial writes from corrupting the chain. The hash covers every field including the previous event's hash — you cannot insert, delete, or modify any event without breaking all subsequent hashes. audit_verify_chain recomputes each hash from scratch rather than trusting stored values.
All data is stored locally. No network requests are made by this server.
| Plan | Price | Included |
|---|---|---|
| Pro | $19/month | Unlimited audit events, full chain verification |
Proprietary — see subscription terms. Source: github.com/Rumblingb/agent-audit-mcp
io.github.ericm1018/skillfm-llm-cost-optimizer-openai-anthropic-usage
io.github.mikerawsonnz/llm-orchestration-agent
io.github.mikerawsonnz/authenticated-llm-agent
labforgedev/copilot-memory-mcp
csoai-org/agent-prompt-injection-firewall-mcp
io.github.mikerawsonnz/authenticated-multi-llm-agent