Horizon Fidelity Monitor
Tracks the structural health of multi-turn conversations by measuring properties LLMs can't see from the inside: semantic drift, information gain, token waste, temporal desync, and causal reachability. Exposes three MCP tools (new_conversation, process_turn, configure_session) that let Claude or Cursor monitor conversation quality in real time and trigger interventions when fidelity drops. Built on information theory and category theory research showing 39% accuracy degradation after five turns in typical agent conversations. Works as a hosted SSE endpoint, a pip-installable library, or a local stdio server. Wraps OpenAI, Anthropic, and LangChain SDKs to inject monitoring into existing agents. Returns 29 signals per turn including circadian cognitive load factors and estimated conversation lifespan before collapse.
- tools
- 3
- auth
- Secret needed
- transport
- SSE
- last updated
- Nov 8, 2026
Tools
3 toolsPublic tool metadata for what this MCP can expose to an agent.
new_conversationReload your MCP client and three tools appear: new_conversation, process_turn, configure_session.
Reload your MCP client and three tools appear: new_conversation, process_turn, configure_session.
No parameter schema in public metadata yet.
process_turn
No parameter schema in public metadata yet.
configure_session
No parameter schema in public metadata yet.
Sponsors
Agent CaffeineSponsoredClose your lid, come back to finished work
SnugSponsoredThe clipboard manager for the AI eraPromote your product hereReach 125k+ visitors monthly
WaxSponsoredBlock anything with supervision, without erasing your iPhonePromote your product hereReach 125k+ visitors monthly
knobsSponsoredSee your site on any phone, any day, in any languageReadme
Horizon Fidelity Monitor
"Quality is not a model property — it is a conversation property."
Horizon is a real-time conversation health monitor for AI agents. It tracks the structural dynamics of multi-turn conversations — semantic drift, information gain, ontological gap width, temporal desynchronisation, circadian cognitive load, conversation velocity, and causal reachability — dimensions that LLMs do not reliably surface from inside the conversation.
Details
- Repository
- github.com/leocelis/horizon
- Website
- leocelis.github.io/horizon
- Category
- Monitoring & Observability
- Transport
- SSE
- Auth
- Required
- Registry
- MCP registry, active
More from leocelis
Related Monitoring & Observability MCP servers
View all
Llm Observability Orchestrationmikerawsonnz/gosce-agentsRun a prompt through a LangChain (system + human) chain over Gemini on Vertex AI; optional LangSmith0stars
Perceptdot Sentryio.github.perceptdot/sentryAI agent vision for Sentry error tracking — with ROI measurement. perceptdot.com0stars
Log Reducerlaunch-it-labs/log-reducerReduces log files for AI consumption — 50-90% token reduction via 18 deterministic transforms.0stars
News Monitor Mcpmalkreide/news-monitor-mcpNews monitoring via WorldNewsAPI: search, sentiment, headlines, trends, alerts (Swiss default)0stars