Skip to content
mcp/harbor
Verified official

AgentWatch MCP

v0.5.1

dev.workers.agentwatch.agentwatch/agentwatch

Read-only watchtower for AI agents on-chain: decoded receipts, plan vs execution, Safe audits.

transport:streamable-http tools:21
Website

Target client

run in your project directory 

claude mcp add --transport http agentwatch https://agentwatch.agentwatch.workers.dev/mcp

setup guides: Claude Code Claude Desktop Cursor VS Code Zed Windsurf Cline Gemini CLI Grok ChatGPT Claude.ai LangChain

Adds it for this project only. Append --scope user to make it available everywhere. Claude Code docs

list_watched Read-only

list watched

get_feed Read-only

get feed

run_audit Mutating

run audit

get_capabilities Read-only

get capabilities

get_alerts Read-only

get alerts

get_pending_proposals Read-only

get pending proposals

get_fidelity Read-only

get fidelity

get_leaderboard Read-only

get leaderboard

get_benchmark Read-only

get benchmark

get_tca Read-only

get tca

register_intent

register intent

add_watch Mutating

add watch

watch_demo_set

watch demo set

refresh_watches

refresh watches

get_agent_passport Read-only

get agent passport

set_agent_passport Mutating

set agent passport

list_recovery_owners Read-only

list recovery owners

set_recovery_owner Mutating

set recovery owner

remove_recovery_owner Mutating

remove recovery owner

get_safe_defaults Read-only

get safe defaults

get_dashboard Read-only

get dashboard

All 21 tools, one page each

2 prompts you invoke yourself

1 resource you attach as context

Mutating and Read-only are read off each tool's name, not its schema — a hint, not a guarantee. Check the server's own documentation before granting access.