Open source MCP
Flow Engine
The context layer AI agents need before touching a codebase.
Flow Engine reads real projects and exposes an MCP to map architecture, locate symbols, generate focused context, measure blast radius, score risk and build refactor plans.
Built by Rodrigo to turn software maintenance into a traceable decision: the agent does not get a file dump, it gets the map, the target, the impact, the risk and the next step.
What it delivers
- Maps areas, entrypoints, boundaries, critical paths and integrations.
- Looks up classes, methods, imports and symbols by name before opening a file.
- Generates Markdown context for LLMs scoped by area, namespace or entrypoint.
- Computes callers, callees, blast radius, change risk and refactor plan per node.
- Supports PHP, TypeScript/JavaScript, Python, Go, Dart and edge extraction from Blade.
Exposed MCP tools
- flow_map
- flow_infra_map
- flow_lookup
- flow_find
- flow_context
- flow_impact
- flow_risk
- flow_refactor_plan
In action
Typical task: plan a change in a 100k+ LOC Laravel + TypeScript monorepo with multiple areas and teams.
flow_mapmaps architecture, areas, entrypoints and boundaries of the project.flow_findlocates candidates by name without opening files.flow_lookupconfirms the exact shape of each candidate straight from the graph.flow_impactreturns callers, callees and blast radius of the target node.
Outcome: In a few minutes, the agent goes from request to plan with the right spot located and the change impact already drawn — no blind file reading.
- Open source
- MCP
- CLI
- Local HTTP
- CI gates
- LLM context
