System Architecture

Complete data flow from AI query to live instrumentation

🤖

External AI

Claude • VS Code • Cursor
⌨️

Coral Ask

Built-in Terminal AI
MCP Protocol
🔄

MCP Proxy

Protocol Bridge
gRPC
🧠

Colony Server

  • MCP Server (Genkit)
  • Service Health & Query
  • RBAC & Audit Logging
  • Telemetry Aggregation (DuckDB)
WireGuard Mesh
🧭

Discovery

👁️

Agents

  • eBPF (Beyla) Metrics
  • OTLP Receiver
  • Telemetry (DuckDB)
  • shell/exec
  • Live Probes
Instrumentation
📦

Application

SDK & Runtime

Architecture Components

⌨️

Coral CLI

The unified command-line interface for all Coral interactions. Includes the Coral Ask terminal assistant and manages the MCP Proxy.

🔄

MCP Proxy

Built into the Coral CLI, this lightweight bridge translates between the Model Context Protocol (stdio) used by AI assistants and the gRPC protocol used by the Colony server.

🧠

Colony

Central coordinator with MCP server, DuckDB storage, and AI orchestration. Provides the unified interface for all AI assistants to query distributed observability data.

🧭

Discovery Service

Global registry that coordinates WireGuard key exchange and NAT traversal, enabling secure agent-to-colony connections across any network.

👁️

Agents

Local observers using eBPF, OTLP, and shell commands to gather telemetry. Run as a sidecar or node agent (DaemonSet) to monitor services with local DuckDB storage.

⚙️

SDK (Optional)

Advanced features like live probes and runtime instrumentation. Enables on-demand eBPF uprobes for deep debugging without redeploying.

Key Architecture Features

🔌 Universal MCP Integration

Colony exposes a standard MCP server that works with any AI assistant: Claude Desktop, VS Code, Cursor, or custom applications. The coral mcp proxy translates between MCP protocol (stdio) and Buf Connect gRPC for type-safe communication.

🔐 Encrypted WireGuard Mesh

All components connect via an encrypted WireGuard mesh coordinated by the Discovery Service. Works across any network boundary with no VPN configuration, firewall rules, or per-environment tooling required.

📊 Edge-Based Observability

Agents run on each service (sidecar or node agent) with local DuckDB storage for recent data (~6hr rolling window). Colony polls agents for summaries, reducing network overhead while maintaining fast queries for detailed telemetry.

🎯 Progressive Integration

Start with Level 0 (zero-config eBPF), optionally add Level 1 (OTLP for OpenTelemetry apps), Level 2 (shell/exec diagnostics), and Level 3 (SDK Live Probes): App owners can integrate a language-specific SDK to launch a runtime bridge with the agent, enabling live probes and deep runtime control.