Architecture Intelligence Platform — Claude Code MCP server
Reconciles static API definitions with live OpenTelemetry traces for AI coding assistants. It surfaces undeclared dependencies before you modify microservice code.
Open on GitHubWhat Architecture Intelligence Platform does
The server bridges declared specifications like AsyncAPI with observed runtime telemetry. Rather than making an agent guess service boundaries, it maps out exact relationships. Dependencies detected in network traffic without documentation receive the OBSERVED_ONLY status, while cataloged endpoints missing from the current traffic window are marked NOT_OBSERVED_IN_WINDOW. Matched relationships become CONFIRMED.
Every generated response ties back to a specific snapshot_id. Using get_evidence, an agent can inspect the exact origin behind any finding, such as an OpenTelemetry span or an AsyncAPI file. When target queue consumers lack clear telemetry proof, the server logs an explicit limitation instead of guessing.
Tools exposed to coding agents operate strictly in read-only mode. Running the packaged demonstration requires Docker with Compose, plus curl and jq. The runtime evaluation pipeline operates without any external LLM keys.
What it helps with
- detect undocumented microservice dependencies from live telemetry
- spot discrepancies between AsyncAPI definitions and runtime traffic
- trace architecture relationship claims back to OpenTelemetry spans
- inspect verified service dependency graphs before refactoring
Example prompts for the agent
- Check architecture drift for order-service using get_architecture_drift
- Fetch the telemetry evidence for LegacyPricingService with get_evidence
- List all confirmed and unobserved dependencies for the fight service
How to install Architecture Intelligence Platform
Install
The MCP registry has no launch command for it: see the repository README
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Architecture Intelligence Platform
How do you set up Architecture Intelligence Platform with Claude Code?
The project is available in the MCP Registry as io.github.michaelegner/architecture-intelligence-platform. You can boot the local environment alongside Neo4j and OpenTelemetry Collector by running examples/runtime-demo/mcp-demo.sh via Docker.
What does Architecture Intelligence Platform do?
It compares declared interface contracts against captured OpenTelemetry traffic to provide verified service topologies. The server highlights unrecorded dependencies and inactive endpoints without fabricating missing data.
Does the platform require an LLM API key to run?
No, the local demo and query engine function without model keys. It processes telemetry traces and contract schemas directly on your machine.
