kicad-mcp-pro (oaslananka) errors

15 recurring problems engineers report in kicad-mcp-pro (oaslananka), ranked by impact.

#23 by errors trackedECAD

Recurring kicad-mcp-pro (oaslananka) errors, by impact

  1. 1A KiCad MCP routing tool hit an obstacle-bypass failure, warned, replied "Routed 2 wire segment(s)" and saved the route anyway, shorting CC1 and USB D- to ground on a USB-C plus STM32 schematic. The MCP's own connectivity graph missed it; only a re-export of KiCad's native netlist caught it.ECAD · Agent failure · 1 report74
  2. 2A KiCad MCP tool placed a THVD1450D transceiver, returned success, and its own component contract check passed 8 of 8 pin-pad numbers plus footprint and 3D. Native KiCad 10.0.7 ERC then reported the placed symbol doesn't match the copy in the library.ECAD · Agent failure · 1 report74
  3. 3PR #1594 fixes "False-PASS #6: empty schematic sheet (0 symbols, 0 wires) or empty PCB (0 footprints) evaluates to NOT_ASSESSED, never to PASS." It also splits schematic symbol rotation, footprint rotation and pick-and-place rotation, which had been mixed. PR #1593 makes sure "nearby wrong-rail capacitors cannot receive electrical PASS" by checking actual pin and net connectivity instead of distance, and labels distance-only checks as advisory. Six numbered false passes in one MCP server. Agents that read PASS from this tool would have shipped boards the check never actually looked at.ECAD · Agent failure · 1 report74
  4. 4PR #1594 fixes "False-PASS #6: empty schematic sheet (0 symbols, 0 wires) or empty PCB (0 footprints) evaluates to NOT_ASSESSED, never to PASS." It also splits schematic symbol rotation, footprint rotation and pick-and-place rotation, which had been mixed. PR #1593 makes sure "nearby wrong-rail capacitors cannot receive electrical PASS" by checking actual pin and net connectivity instead of distance, and labels distance-only checks as advisory. Six numbered false passes in one MCP server. Agents that read PASS from this tool would have shipped boards the check never actually looked at.ECAD · Agent failure · 1 report74
  5. 5Someone evaluating KiCad MCP servers read the project gate code. It ties done to kicad-cli DRC, but a board with unfilled or stale zones can pass while copper is missing, and unconnected items do not fail the gate. route_single 'returns a fixed success message without reading the result back.'ECAD · Agent failure · 1 report74
  6. 6The KiCad IPC companion treated a JSON-RPC error, a response with no result, and ok=false as successes, so it could report PCB context synced when the MCP operation had failed.ECAD · Agent failure · 1 report74
  7. 7The schematic writer scans for (wire at every character, including inside quoted strings, so if a symbol value or text item contains a wire expression, the next write empties that string and appends the embedded segment as a real wire, which can join nets. The tool reports success. The structural-loss guard compares node counts, and the wire count goes up, not down, so it passes. A single sch_add_component call is enough to trigger it.ECAD · Agent failure · 1 report74
  8. 8Follow-up fix to #1518: the PCB gate now blocks when the DRC report has no unconnected list, ignores connectivity, or the stored zone fill differs from a refill on a throwaway copy. Before this, those cases could read as PASS.ECAD · Agent failure · 1 report66
  9. 9On a real KiCad 10.0.8 board, ERC and DRC passed and placement scored 100 out of 100, yet the overall quality gate still returned FAIL. It labelled a cosmetic advisory as an error, said 'Blocking issues remain', and pointed the agent at the warning as the first thing to fix. When a gate's verdict and its findings disagree, the agent chases the wrong fix.ECAD · Agent failure · 1 report66
  10. 10The documented readiness command for kicad-mcp-pro exits with ModuleNotFoundError, and the PR says it fixes the launch but does not meet real-provider or scored board-attempt acceptance.ECAD · Agent failure · 1 report66
  11. 11The JLCPCB annular ring check treated any off-center PTH drill as unmeasured even though KiCad states the offset exactly, and it now computes the narrowest ring and fails closed on oval or odd drills.ECAD · Agent failure · 1 report66
  12. 12The pin and gate swap tools validated the request and saved an intent file but never changed the schematic, and nothing ever finished the back-annotation. The parity matrix had called it covered. Now it's partial and coverage drops from 85.7% to 84.4%. In the companion PR, moving a rotated footprint without passing a rotation reset it to 0 degrees.ECAD · Agent failure · 1 report66
  13. 13The pin and gate swap tools validated the request and saved an intent file but never changed the schematic, and nothing ever finished the back-annotation. The parity matrix had called it covered. Now it's partial and coverage drops from 85.7% to 84.4%. In the companion PR, moving a rotated footprint without passing a rotation reset it to 0 degrees.ECAD · Agent failure · 1 report66
  14. 14Third fix from #1518 today: the manual track tools used to tell the agent routing was complete just because the writer returned; they now say the write is unverified and point to DRC and the quality gate.ECAD · Agent failure · 1 report66
  15. 15Tools named pcb_add_track and pcb_route_trace only place one straight segment, while the names promised path planning, obstacle avoidance, differential pairs, and DRC that the tools do not do.ECAD · Agent failure · 1 report66

How it fails

  1. Agent failure15

Engineering tools that address these errors

Sourced from public community reports. Costs and delays on individual entries are modeled estimates, not measurements.