Bring Any AI (MCP)
Goal: Connect any AI to your workshop - to read and to build.
OpenX is one MCP server, and any MCP-native AI can drive it. MCP (the Model Context Protocol) is the open standard for connecting AI to data and tools - HTTP for AI. Because it is a standard, your connections are yours: fire us, keep running. Azi uses this same server; so can the agent you bring, including a local model for a fully owned loop.
Your MCP server URL (shown in APIs -> MCP Configuration):
https://openx.fathym.com/oi-api/mcp
Install with fai (fastest)
fai mcp install https://openx.fathym.com/oi-api/mcp --name openx --auth "YOUR_TOKEN"
Registers OpenX with Claude Code, Cursor, Cline, and other detected agents at once. Browse the tools
with fai mcp preview openx. For a desktop app, add the same URL + token to its MCP config.
Set up your AI
fai mcp install configures every coding agent it detects on your machine in one go - Claude Code,
Cursor, Cline, VS Code (Copilot), Windsurf, JetBrains AI, Goose, Codex, Gemini, and more. (The Claude
Desktop app uses a small bridge - see its page.) Prefer to wire one up by hand, or use a no-code tool?
Pick your client:
- Claude (Code & Desktop)
- Cursor & VS Code
- Custom clients - build your own over JSON-RPC
- No-code apps - Dify, Langflow, Flowise, n8n, Make
Read from any AI
| Tool | Does |
|---|---|
list_data_connections | Discover your data sources |
run_warm_query | Run a named source of truth |
run_adhoc_warm_query | Try a one-off query |
tail_telemetry | Watch live data |
get_workspace_flow | See the current workshop + pending proposals |
The warm_query_catalog resource lists your queries. Ask any connected AI "what are the average
signups this hour?" and it calls run_warm_query against the same source of truth.
Build from any AI
HOW BUILDING WORKS
Propose over MCP, accept in the portal
Any AI can also build: it calls manage_proposals to propose a new connection, surface, or warm
query. Proposals land in your OpenX workspace tagged by origin (you or an AI). You accept them in
the portal, then deploy. There is no accept tool over MCP by design - building is open to any AI,
but what goes live is always your call.
App-makers
Dify, Langflow, Flowise, n8n, and Make speak MCP (or call the REST endpoints) - point them at the same URL with your token - see No-code apps.
Next steps
- Crystallize more sources of truth -> Warm Queries
- Package and share your own workbench -> Host Your Own Workbench