Engineering case study
Extending an Autonomous Software Factory to MCP
How MiniMindsLab evolved from an autonomous browser-tool factory into a multi-target software system that selectively recompiles production capabilities for Model Context Protocol.
The problem
MiniMindsLab already discovered, specified, generated, repaired, browser-tested, committed, deployed, and live-verified web tools. The next question was whether a proven capability could become something other software could call without duplicating the entire product factory.
The architecture
The factory was split at the delivery boundary. Product discovery and the frozen capability contract remain upstream. After a canonical web build exists, a delivery-target decision determines whether the capability stops at the web or continues into another implementation factory.
The MCP factory
MCP-routed capabilities are reimplemented as deterministic modules, tested through an actual Model Context Protocol server/client handshake and tools/call sequence, committed into an immutable MCP canonical index, then assembled into a multi-tool runtime on mcp.minimindslab.com.
Production proof
Conference Badge Readability Tester established the runtime precedent. JSON Schema vs Sample Payload Diff Checker and JSON Contract Compatibility Checker then proved that Product Brain-originated capabilities could travel from autonomous discovery through the web factory and selectively continue into the MCP factory and public developer catalog.
One source of truth per surface
The public MiniMinds web directory is derived from the canonical web tool index. The MCP runtime, machine catalog, Registry publication, and human developer catalog are derived from the canonical MCP tool index. The sites do not maintain a separate manual list of MCP capabilities.
The result
MiniMindsLab is now a multi-target software factory rather than only a web-tool generator. A capability can be discovered once, proven on the web, evaluated for target fit, and compiled into additional delivery surfaces without weakening the independent QA and canonicalization requirements of each target.