HorusComparison
Horus vs MCP: raw access is not the same as grounded investigation context
MCP gives AI agents tool access. Horus turns logs, code, queues, metrics, deploys, and evidence into grounded investigation context agents can reason over.
MCP (the Model Context Protocol) gives an AI agent a standard way to call tools and data sources. Horus is not an MCP competitor — it's the investigation layer behind those tools that turns raw signals into context an agent can actually reason over.
MCP is useful — but raw tool access isn't enough
MCP solved a real problem: one interface for an agent to reach a database, a search index, or a file system. But access is not understanding. An agent with ten MCP servers still has to rediscover your system every session — which service owns checkout, where the queue backs up, what changed in the last deploy.
- MCP exposes tools; it doesn't decide which signals matter for a given incident.
- Each server returns raw output — logs here, metrics there — with no shared model linking them.
- The agent pays the context cost of stitching it together, every single time.
Why agents burn context with scattered MCP servers
- No memory: each session re-fetches and re-derives the same structure.
- No correlation: a log line, a slow query, and a recent commit arrive as unrelated blobs.
- No prioritization: the agent reads everything to find the few things that matter.
What Horus adds on top
Horus runs locally in your repo, connects your runtime sources, and produces a structured, evidence-backed investigation — not a pile of tool output.
- Project memory — a code graph plus service/queue map so the system model persists across sessions.
- Evidence — logs, metrics, queue and database state captured and linked to the code they implicate.
- Timeline — what changed and when, correlated with when symptoms began.
- Ownership — which symbols, files, and services are involved.
- Hypotheses and next checks — a deterministic report to act on, not guess from.
The one-line difference
MCP connects agents to tools. Horus connects agents to an investigation model.
When to use each
| Reach for | When you need |
|---|---|
| MCP | A standard way for an agent to call a specific tool or data source |
| Horus | Scattered signals turned into investigation-ready context across code, logs, queues, metrics, deploys, and ownership |
Horus complements MCP — it doesn't replace it
These layers stack. Keep your MCP servers for tool access; let Horus do the investigation and hand the agent grounded context. Horus can even be the source of investigation context an agent reaches — including over MCP.
See the docs, the investigation walkthrough, or how Horus reads runtime connectors and source intelligence.