📡 MCP Radar

Nouman-Amjad/Cairn

Agentic incident-analysis copilot: MCP tool servers, a cost-aware model router, an approval-gated write path, and a companion npm CLI for platform analysis, a live dashboard, and log chat.

5 stars
3 forks
Python
momentum ▲ 14.3
created 2026-08-21
on radar since 2026-08-23
agentic-aiclidashboarddevopsdevops-toolsincident-managementincident-responsekubernetesllm-agentlog-analysismcpobservabilityplatform-engineeringsre
View on GitHub ↗Homepage ↗

About Cairn

An agentic incident-analysis copilot. Ask "why did checkout latency spike at 3am?" in plain English; Cairn queries the observability stack, correlates against the deploy timeline, retrieves the relevant runbook, and proposes a root cause with evidence — and can execute a remediation, behind a human approval gate.

This repository implements docs/architecture.md.

The companion npm package (npx @nouman-amjad/cairn dashboard) serves a live local dashboard over any project — no build step, no dependencies, no network.

The four commitments, and where they live in the code

Commitment Where it is enforced ------ Tools are the product. Every capability is an MCP tool; the agent has no privileged path to any backend. services/cairn-mcp-, packages/cairn-mcp-kit Inference is a routed resource. A cost-and-sensitivity router splits work between a local 8B and a frontier model. routing.py Write actions are never autonomous. A durable approval state machine with idempotency keys and an append-only audit log. service.py The eval harness is a first-class service. 30 scenarios with ground-truth causes and 7 gated metrics. services/cairn-eval

cairn-deploy and cairn-infra are vendored here for review. In production they are separate repositories with separate lifecycles — see ADR-013.

No GPU and no API key are needed for any of the above. make eval runs in heuristic mode, which exercises the whole pipeline — loop, tool capping, persistence, the seven metrics, the gate — with a scripted stand-in instead of a model. See the caveats.

From the project README.

Maintaining this server?

Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:

[![On MCP Radar](https://mcp.liqiwa.com/badge.svg)](https://mcp.liqiwa.com/s/Nouman-Amjad--Cairn.html)

Related MCP servers

deadtrickster/serenedash

A live terminal dashboard for a SereneDB server: storage, memory, sessions, per-thread CPU and a perf-backed profile, cheap enough to leave open.

⭐ 4Python
SujalXplores/Agent-K

Code-enforced incident-response agent that investigates SigNoz alerts, publishes evidence-backed root-cause claims, and gates every action behind a safety policy. Built for Agents of SigNoz hackathon.

⭐ 3Python
Resetnak/cooldeck

Keyboard-first terminal dashboard for Coolify: deployments, live logs across your whole fleet, and an MCP server for your agent

⭐ 9Go
0xwilliamortiz/openclaude-improved

runs anywhere. uses anything

⭐ 563TypeScript
bar181/bar-observatory

Deterministic, local-only audit reports for Claude Code AI agent sessions. Rust + SQLite. Zero network calls. MCP server for agents.

⭐ 18HTML
priyansh-narang2308/TheBlacklisted-NitroStack

Protocol-0 is an MCP (Model Context Protocol) server that gives AI assistants, Claude, Cursor, ChatGPT, NitroChat, or any MCP-compatible client, real authority to act as your Site Reliability Engineer

⭐ 4TypeScript

📬 Get the weekly radar in your inbox

The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.