rfix.ai

RFIX MCP server-side mode helps agents plan RF scenarios, search the RFIX block catalog, and export launchable `.rfix.gz` projects. It does NOT generate IQ data and does NOT talk to rfix-api or rfix_python. WORKFLOW (follow this order): 1. Call `rfix_plan_scenario(description=...)` — returns block schemas, composition rules, constraints, and a reference ReactFlow/editor-project example. 2. COMPOSE YOUR OWN GRAPH: Use the reference graph as inspiration, but adapt it to the user's request. Add/remove nodes, adjust parameters, and change edges. 3. Call `rfix_export_rfix_project(payload=<your composed graph/project>)` — writes a `.rfix.gz` project and returns launch metadata for desktop and web. That is the happy path for server-side launch flows. Do NOT search the filesystem for RFIX examples. ACCEPTED EXPORT INPUT SHAPES: - GCF/LLM output: `{reactFlowNodes, reactFlowEdges, gridGlobalSettings}` - Exported graph: `{nodes, connections, globalSettings}` - Editor/.rfix-style project JSON - Backend payloads are also accepted and will be reverse-transformed. CRITICAL RULES: - Modifier/channel blocks MUST connect to an upstream signal source. Never create isolated modifiers. - Signals cannot connect directly to other signals — use modifiers between them. - All edge endpoints and any table block references MUST point at existing node IDs. - three_d_emitter and three_d_sensor MUST be connected through a 3D channel (three_d_channel_los/rayleigh/rician). - `rfix_export_rfix_project` returns `launch_grant`, `desktop_deeplink`, `web_launch_url`, `web_embed_url`, and `download_desktop_url`. Read and report those exact values.