Two issues with run_erc on KiCad 9:
1. kicad-cli returns non-zero exit code when ERC violations exist.
The handler treated this as a command failure and returned early
with success=false, even though valid JSON output was produced.
Fix: check for output file existence instead of exit code.
2. KiCad 9 nests violations under sheets[].violations instead of
(or in addition to) the top-level violations[] array used by
KiCad 8. The handler only read the top-level array, reporting
0 violations on schematics with sub-sheets.
Fix: iterate sheets[] and collect all nested violations.
Both fixes are backward-compatible with KiCad 8.
6 unit tests added covering non-zero exit codes, KiCad 8 top-level
violations, KiCad 9 sheets[] nesting, mixed structures, zero
violations, and missing output files.
export_netlist was returning "Unknown command" because no Python handler
existed. generate_netlist was timing out (30s) due to an O(nets × components
× pins) wire-graph algorithm with a new PinLocator instantiated per net.
Both handlers now delegate to `kicad-cli sch export netlist`:
- export_netlist: new handler; writes KiCad XML / Spice / Cadstar / OrcadPCB2
to the caller-supplied outputPath. Added schematicPath parameter to the TS
tool definition (was absent, making file export impossible).
- generate_netlist: replaces the slow wire-graph with kicad-cli + XML parse;
returns the same {components, nets} JSON the TS handler already expected.
Also adds _find_kicad_cli_static() so both handlers share CLI discovery
without depending on ExportCommands (which requires a loaded pcbnew board).
Cleaned up generate_netlist schema in tool_schemas.py (removed outputPath and
format fields the handler never used), updated MCP tool descriptions and
SCHEMATIC_TOOLS_REFERENCE.md to clearly distinguish the two tools.
26 unit tests added covering parameter validation, subprocess mocking,
format mapping, XML→JSON parsing, and error/timeout propagation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Half of all valid KiCAD schematic pin positions are on the 50-mil
(1.27mm) grid but not the 100-mil (2.54mm) grid — e.g. 26.67mm = 21 ×
1.27mm. Snapping to 2.54mm displaced those coordinates by 1.27mm,
moving labels off their pins and increasing floating-label count.
KiCAD source confirms: DEFAULT_CONNECTION_GRID_MILS = 50 and the ERC
off-grid check uses exact integer modulo against this value, so any
displacement breaks connectivity unconditionally.
Also update the kicad-source absolute path in CLAUDE.md.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
get_pin_net was a superset of get_wire_connections with the same
coordinate-based flood-fill but two extra response fields (net, query_point)
and a reference+pin input mode. Having both tools confused LLM tool selection.
get_wire_connections now:
- Returns net (label name or null) and query_point in all response paths
- Accepts reference+pin input in addition to x/y coordinates,
resolving the pin endpoint via PinLocator internally
get_pin_net tool, handler, schema, TS registration, and tests removed.
test_wire_connectivity.py updated with coverage for all new behaviour.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Answers "what net is pin X of component Y on?" without requiring
callers to triangulate from list_schematic_nets or know a wire
coordinate first.
Accepts either {reference, pin} (resolved via PinLocator) or {x, y}
coordinate. Returns net label name (or null for unnamed nets), all
connected pins, wire segments, and the resolved query point.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds a new MCP tool that snaps wire endpoints, junction positions, and
net label coordinates to the nearest grid point (default 2.54 mm). Off-grid
coordinates cause wires that appear visually connected to fail ERC checks
because KiCAD uses exact IU integer matching internally; this tool eliminates
that class of error before running ERC.
- python/commands/schematic_snap.py: core snap logic with in-place sexp mutation
- python/kicad_interface.py: route + handler
- python/schemas/tool_schemas.py: JSON schema (gridSize, elements params)
- src/tools/schematic.ts: TypeScript MCP tool registration
- tests/test_snap_to_grid.py: 20 unit + integration tests (all passing)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Detects wire segments with at least one dangling endpoint — an endpoint
not connected to a component pin, net label, or another wire. These
cause ERC 'wire end unconnected' violations and are a common symptom of
incomplete routing or stray stub wires.
Algorithm uses exact KiCad IU (10 000 IU/mm) coordinate matching,
consistent with wire_connectivity.py:
1. Build an endpoint-frequency map for all wires (IU precision)
2. Collect anchored IU points: component pins (via PinLocator),
net labels / global_labels, power symbol pins
(via _parse_virtual_connections)
3. An endpoint is dangling when it is touched by exactly one wire AND
is not an anchored point; the containing wire is reported
Does not require the KiCad UI to be running.
Changes:
python/commands/schematic_analysis.py — find_orphaned_wires() function
python/kicad_interface.py — handler + route registration
python/schemas/tool_schemas.py — MCP schema entry
src/tools/schematic.ts — TypeScript server.tool() call
tests/test_schematic_analysis.py — 7 integration tests
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
add_schematic_net_label now accepts optional componentRef + pinNumber to
snap the label directly to the exact pin endpoint via PinLocator, removing
all approximation risk. The response always includes actual_position and,
when snapping was used, snapped_to_pin — so the caller gets confirmation
of exactly where the label landed.
connect_to_net return type changed from bool to Dict, returning
pin_location, label_location, and wire_stub on success so agents no
longer need a separate verification call to confirm placement.
connect_passthrough updated to check result.get("success") against the
new dict return. tool_schemas.py and schematic.ts updated to match
(position is now optional, componentRef/pinNumber/labelType/orientation
added, connect_to_net schema field names corrected).
17 new unit tests in tests/test_net_label_pin_snapping.py.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Move all Python test files from python/tests/ to the top-level tests/
directory to match the project's CONTRIBUTING.md guidelines. Update
sys.path inserts and template path references to reflect the new
location.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously passing mirror='x' or mirror='y' to rotate_schematic_component
always raised:
'NoneType' object has no attribute 'value'
Root cause: kicad-skip has no API for setting (mirror x/y) on a placed
symbol instance. The handler tried to use a non-existent kicad-skip
attribute, returning None, then calling .value on it.
Fix: add _apply_mirror_to_symbol_sexp() which directly patches the
(mirror x/y) S-expression token in the .kicad_sch file. The rotate
handler now applies rotation via kicad-skip (which works fine) and
delegates mirror to this helper.
The helper:
- Inserts (mirror x/y) immediately after the (at x y rot) token
- Removes any existing mirror token before inserting the new one
(prevents duplicate tokens when toggling axis)
- Returns False gracefully when the reference is not found
Tests: 6 unit tests in tests/test_rotate_schematic_mirror.py
covering: add x, add y, remove, replace, unknown ref, no-mirror smoke test.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously the path argument to create_schematic() was accepted by the
tool schema but silently ignored in SchematicManager.create_schematic()
(python/commands/schematic.py). The schematic file was always written
using only the bare filename, resolving to the MCP server's working
directory.
On systems where that directory is not writable (e.g. Program Files,
OneDrive-synced folders) this caused:
[Errno 13] Permission denied: 'myproject.kicad_sch'
Fix:
- Add path: Optional[str] = None parameter to SchematicManager.create_schematic()
- Compute output_path as os.path.join(path, base_name) when path is given
- Forward the resolved path from _handle_create_schematic() into create_schematic()
Tests: added tests/test_create_schematic_path.py with three unit tests
covering: path respected, no-path fallback, and no double-suffix on .kicad_sch names.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The PR added a second server.tool("move_schematic_component", ...) at line 1127
without removing the original registration at line 722, causing the server to
fail on startup with "Tool move_schematic_component is already registered".
Also adds tests/test_ts_tool_registry.py which scans all src/tools/**/*.ts files
for duplicate server.tool() names so this class of bug is caught automatically.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add isort configuration (profile=black, line_length=100) to pyproject.toml,
add isort pre-commit hook, and auto-sort imports across all Python source files.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add [tool.black] config to pyproject.toml and Black hook to
.pre-commit-config.yaml (rev 26.3.1), then auto-format all Python
source and test files with line-length=100, target-version=py310.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
