Overview
What the Axon MCP Server does, why it exists, and what you can build with it.
What it is
The Axon MCP Server is a Model Context Protocol server that gives AI assistants expert-level access to an Axon codebase. It indexes Axon source files and HTML documentation, exposes search and code-generation tools over MCP, and can execute queries against a live SkySpark instance.
Axon is SkySpark's domain-specific language for building automation, HVAC control, and energy management. The MCP server turns a folder of Axon code into a structured knowledge base that any MCP-compatible client — Claude Desktop, Cursor, the MCP CLI — can query.
What you can do with it
- Search across Axon code and docs — FlexSearch-powered full-text search with relevance scoring, operator-level queries, and regex patterns.
- Generate new Axon code — 33 curated YAML templates for common HVAC, energy, fault, and data workflows. Generated code is validated semantically before being returned.
- Execute code against a live SkySpark instance — run Haystack queries, evaluate Axon expressions, and commit functions with automatic versioning.
- Analyse call graphs — find callers and callees, detect circular dependencies, identify the most-called or most-complex functions.
- Semantic search with vector embeddings — natural-language queries that return code by meaning, not just keyword match.
How it fits together
┌──────────────────┐ MCP ┌──────────────────────┐ HTTP ┌────────────┐
│ AI client │ ─────────────▶│ Axon MCP Server │ ────────────▶│ SkySpark │
│ (Claude, Cursor) │ │ • search index │ │ (Haystack) │
│ │ │ • code generator │ └────────────┘
│ │ ◀─────────────│ • graph analyser │
└──────────────────┘ results │ • OAuth 2.1 │
└──────────────────────┘
│
▼
.axon / .trio files
HTML documentation
Connections run over stdio (for local AI clients that spawn the server as a subprocess) or http (for remote clients behind OAuth). The dashboard at /admin provides a web UI for sessions and user management.
Key features
- Fast search — ~50 ms full-text across thousands of HTML files; initial indexing takes 30–60 s and is cached.
- OAuth 2.1 with PKCE — multi-user authentication, scoped tokens (
mcp:read,mcp:write,mcp:admin). - Function versioning — auto-keeps the last 4 revisions of every function committed via the server.
- Auto-discovery — configure one SkySpark instance; the server enumerates projects and syncs their functions automatically.
- Local SQLite persistence — user store, backup manifests, and graph metrics all live in a single file via Prisma.
Project status
The server is in active development and is considered stable for search, execution, and code-generation workflows. License: ISC. Issues and contributions are welcome on GitHub.