Causely Open Source is where we share the tools behind our causal context layer for AI ops agents: MCP skills, benchmarks, and the synthetic environments we use to test agent accuracy against ground truth. Causely helps agents resolve issues faster, burn fewer tokens, and stop hallucinating.
The Causely MCP server is a hosted endpoint, always current, no repo required:
https://api.causely.app/mcp
Works with Claude Code, Claude Desktop, Cursor, or any MCP-compatible client. Full setup instructions and tool reference: docs.causely.ai/agent-integration/
A few of the things we're working on or sharing:
- causely-client: MCP Skills and plugins for Claude Code, Claude Desktop, and Cursor. Pre-built investigation workflows (alert triage, postmortems, blast radius) for the Causely causal reasoning MCP server. Fork it to build the tool-selection logic for your own reliability agent.
- awesome-synthetic-apps: Demo applications and telemetry generators for testing agent accuracy against known failure scenarios.
- chaosmania: Ground-truth fault injection for microservices, used to benchmark whether an agent's diagnosis actually matches the real root cause.
- automatic-instrumentation-lab: Techniques for automatic instrumentation, groundwork for any observability pipeline feeding an agent.
Across 72 experiments spanning four agent frameworks, adding Causely's causal model to agent workflows produced:
- 100% fault scenario accuracy, up from 75% without causal context
- 48% fewer tokens per investigation on average
- 63% faster mean query time
Full methodology: Benchmark or reproduce it with this repo.
We're building in the open because we believe in community-driven innovation. Want to get involved?
- 📥 Clone & explore the code
- 🐞 Open issues or suggest enhancements
- 🔁 Submit PRs, we welcome contributions
- 💬 Start a discussion, we're listening
🛠️ This organization is maintained by the team at Causely.
