Move wire connectivity logic from _handle_get_wire_connections into
commands/wire_connectivity.py. Use KiCad's internal integer unit system
(10,000 IU/mm) with exact coordinate matching instead of tolerance-based
float comparison, mirroring how KiCad itself determines connectivity.
Key improvements:
- Exact integer matching for wire endpoints (O(1) dict lookup vs O(n) grid scan)
- Junction support for T-connections
- Multi-unit symbol support (removed incorrect processed_refs dedup)
- Single public API: get_wire_connections()
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Library symbols use y-up coordinates while schematics use y-down. The
_transform_local_point function was not negating y, causing asymmetric
symbols (e.g. power:VEE) to have their bounding boxes computed in the
wrong direction — missing overlaps with adjacent components.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
KiCad's native ERC already checks for unconnected pins with better
accuracy (hierarchical sheets, bus connections, custom rules). Remove
the reimplemented version and its dead helper _parse_no_connects.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Parse graphical elements (rectangle, polyline, circle, arc, bezier) from
lib_symbols definitions to compute accurate symbol bounding boxes instead
of relying on pin positions with hardcoded degenerate expansion. This
fixes bbox accuracy for ICs (previously too small), tiny 2-pin passives
(previously too large), and single-pin symbols.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Eliminate repeated file parsing by extracting _extract_lib_symbols helper
that walks already-parsed sexp_data once instead of re-reading the file
per symbol via PinLocator. Support diagonal wire overlap detection using
cross-product parallelism and 1D projection. Fix wire region inclusion to
use AABB intersection for pass-through wires. Normalize view region
coordinates. Clarify tolerance docstrings across Python, TS, and schema.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Wires that start at a component pin but continue through the body were
incorrectly suppressed as "valid connections." Now nudges the pin endpoint
toward the other end and re-tests intersection — if the shortened segment
still hits the bbox, the wire passes through and is flagged.
Renamed the tool from check_wire_collisions to find_wires_crossing_symbols
across all layers (Python, handler, schema, TypeScript) to clarify that it
finds wires crossing over component symbols, which is unacceptable in
schematics.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The overlap detection was comparing center-to-center Euclidean distance with
a 0.5mm tolerance, missing components whose bodies physically overlap but have
different centers (e.g. a resistor placed inside an opamp triangle). Now uses
AABB intersection on pin-derived bounding boxes, matching the approach already
used by check_wire_collisions. Extracted shared bbox logic into
_compute_symbol_bbox_direct and _aabb_overlap helpers.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Previously, the endpoint suppression logic skipped any wire where at least
one endpoint touched a pin, hiding the case where both endpoints are pins
of the same component (a direct pin-to-pin short through the body).
Replace the single endpoint_matches_pin loop with separate start_at_pin /
end_at_pin checks; suppress only when exactly one endpoint is at a pin.
Add regression test to cover this case.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The same reference-lookup bug fixed in check_wire_collisions (where
PinLocator.get_all_symbol_pins always resolves to the first symbol
with a given reference) also affected:
- find_unconnected_pins: would check wrong pin positions for all
unannotated components after the first, producing false connected/
unconnected reports.
- get_elements_in_region: would return wrong pin coordinates for
unannotated components in the queried region.
Both now use _compute_pin_positions_direct (fetching pin defs by
lib_id and applying each symbol's own position/rotation/mirror),
matching the fix already applied to check_wire_collisions.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
PinLocator.get_all_symbol_pins resolves symbols by reference designator,
so when multiple components share the same unannotated reference (e.g. "Q?"),
it always returned the first match's pin positions. Every duplicate then
got an identical bounding box, causing a single wire to be flagged against
all N instances instead of only the ones it actually crosses.
Fix: add _compute_pin_positions_direct() that computes absolute pin positions
directly from each symbol's own (at x y rotation) and (mirror ...) data plus
pin definitions fetched by lib_id — no reference-name lookup involved.
Also extend _parse_symbols to capture mirror_x/mirror_y flags.
Add regression test: two "R?" at different positions, wire crossing only
one → must produce 0 collisions against the far-away component.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add five new read-only schematic analysis MCP tools:
- get_schematic_view_region: export cropped schematic region as PNG/SVG
- find_unconnected_pins: list pins with no wire/label/power connection
- find_overlapping_elements: detect duplicate symbols, stacked labels, collinear wire overlaps
- get_elements_in_region: list all symbols/wires/labels in a bounding box
- check_wire_collisions: detect wires passing through component bodies
Includes Python handler dispatch, tool schemas, TypeScript server bindings,
the schematic_analysis command module, and a full test suite (28 tests passing).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Auto-detects runtime: Java 21+ direct or Docker/Podman fallback.
Discovers docker/podman via PATH instead of hardcoding.
README includes one-time setup with JAR download + Docker pull.
31 tests covering both execution modes.
4 new MCP tools: autoroute (full DSN→Freerouting→SES pipeline),
export_dsn, import_ses, check_freerouting. Requires Java 11+ and
freerouting.jar. Includes 21 test cases and README usage examples.
- Add python/tests/conftest.py with MagicMock stubs for pcbnew/skip
- Add python/tests/test_schematic_tools.py with 29 tests covering
WireManager.delete_wire, WireManager.delete_label (unit + integration),
and parameter validation for all 11 new _handle_* methods
- Apply black formatting to component_schematic.py, wire_manager.py,
and kicad_interface.py
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
delete_schematic_net_label was using kicad-skip's write() which silently
discards in-memory _elements mutations. Replaced with WireManager.delete_label()
that uses the same sexpdata round-trip approach already used by delete_schematic_wire.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
kicad-skip's write() serializes from original parsed data, silently
discarding in-memory _elements mutations. Switch to the same sexpdata
approach used by add_wire: parse, find matching wire, delete the entry,
write back. Also adds WireManager.delete_wire() static method.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
ElementCollection in kicad-skip doesn't implement remove(), causing
'WireCollection object has no attribute remove' errors. Access the
underlying _elements list directly for wire, label, and symbol deletion.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Detects F.Cu<->B.Cu layer mismatch in route_pad_to_pad
- Splits route into: trace on start layer + via at midpoint + trace on end layer
- route_trace description warns: use route_pad_to_pad for cross-layer routing
- route_pad_to_pad description highlights automatic via insertion
Bug 1 — Rounded corners missing (src/tools/board.ts schema)
Symptom: Claude sends shape='rectangle' + radius=2.5, but the Zod schema
only listed 'rectangle', 'circle', 'polygon' as valid shapes. The JS
handler never passed cornerRadius through the IPC path, and Python only
generated arcs for shape='rounded_rectangle'. Result: 4 straight lines,
no arcs.
Fix: Added 'rounded_rectangle' and 'cornerRadius' to the Zod schema in
src/tools/board.ts so Claude can send the correct shape directly.
Also added a Python-side auto-upgrade: if shape='rectangle' and
cornerRadius>0, silently promote to 'rounded_rectangle' so legacy
callers still work (python/commands/board/outline.py).
Bug 2 — Board outline placed at (-w/2, -h/2) instead of (0, 0)
Symptom: A 30x30 mm board outline was placed centred at the origin
(start -15 -15) ... (end 15 15) instead of the expected top-left at
(0,0) → (30,30).
Root cause: src/tools/board.ts extracted x/y from params and renamed
them to centerX/centerY before forwarding to Python:
const { x, y, ...otherParams } = params;
callKicadScript('add_board_outline', { centerX: x, centerY: y, ...otherParams })
Python received centerX=0, centerY=0 and used that as the board centre,
placing the outline at (-15,-15)→(+15,+15).
Fix: Pass x/y directly as top-left corner coordinates. Python already
contains the correct logic: center = x + width/2, y + height/2. Removed
the incorrect rename in board.ts:
callKicadScript('add_board_outline', { shape, ...params })
Bug 3 — print() on stdout corrupted JSON-RPC protocol → Timeout
Symptom: Claude Desktop reported 'Timeout' for add_board_outline even
though Python completed the command in <5 ms. KiCAD was falsely launched.
Root cause: Debug print() statements written to stdout (the same channel
used for JSON-RPC responses) injected non-JSON lines into the protocol.
Node.js could not parse the response → timeout → Claude Desktop invented
a 'KiCAD UI required' workaround.
Fix: Removed all print() calls from outline.py. Logger.info/debug writes
to the log file (~/.kicad-mcp/logs/kicad_interface.log) and is safe.
Files changed:
src/tools/board.ts — Zod schema + removed centerX/centerY rename
python/commands/board/outline.py — auto-upgrade rectangle+radius, remove print()
- searchTools() previously only searched routed tool categories, so
direct tools like snapshot_project, save_project, create_project
returned 0 results when Claude called search_tools('snapshot')
- Direct tools are now searched first, with a hint that they must be
called directly (not via execute_tool)
Also includes routing.py zone-inset fix for rounded board corners
(already staged from previous session)
Claude sends {shape:'rectangle', params:{x,y,width,height,...}} but handlers
were reading params.get('width') on the outer dict → None → wrong board size.
outline.py: extract inner = params.get('params', params) at method start,
read all dimensions from inner dict.
kicad_interface.py (_ipc_add_board_outline): delegate 'rectangle' to SWIG
path (same as 'rounded_rectangle') so Claude's shape+dims call is handled
correctly. For polygon IPC path: also unwrap inner params for points/width.
FootprintLoad+Flip() on standalone footprint (not yet in board) causes
a 30-second block in KiCAD 9 pcbnew. Solution: add footprint to board
first, then call Flip() with board context available.
Fixes B.Cu placement timeout that killed the Python process.
create_component_instance used line-based insertion which placed new
symbols BEFORE (kicad_sch ...) header when the file was written as
a single line by sexpdata.dumps(). Switch to rfind()-based string
insertion which is format-independent.
Also remove StreamHandler(sys.stdout) from logging — Python logs now
go only to file (~/.kicad-mcp/logs/kicad_interface.log) to avoid
polluting MCP stderr with INFO/DEBUG entries shown as [error].
- add_schematic_net_label: warn in description that coords must be exact pin endpoints; recommend connect_to_net instead
- connect_to_net: stub wire direction now follows pin angle (was hardcoded +X)
- pin_locator.py: add get_pin_angle() and _get_lib_id() helpers
- new tool: get_schematic_pin_locations(schematicPath, reference) → returns exact x/y of every pin endpoint, so Claude can place labels correctly
KiCad 9 refuses to load any schematic whose lib_symbols section contains
an (extends ...) clause, because the extends mechanism is only valid
inside .kicad_sym library files - not inside schematics.
The previous implementation kept the (extends ...) clause and prepended
the parent symbol block with a qualified name, but left the child's
extends referring to the unqualified parent name, causing:
'Error loading schematic: No parent for extended symbol Q_NMOS_GSD'
Fix: add _iter_top_level_items() and _inline_extends_symbol() helpers.
When a library symbol uses extends, the parent's content (pins, graphics,
sub-symbols) is now merged directly into the child definition:
- Parent properties are overridden by child property values
- Sub-symbol names are renamed from ParentName_X_Y to ChildName_X_Y
- The (extends ...) clause is removed entirely
- Only the fully-resolved child symbol is injected into lib_symbols
Affected symbols include Transistor_FET:2N7002, Transistor_BJT:*,
Regulator_Linear:*, Regulator_Switching:* and many others (~30% of the
KiCad standard library uses extends).
Fixes#52
All mounting holes were assigned the same reference "MH", violating
PCB conventions and causing conflicts. Now queries existing footprints
to find the next available MH number.
kicad-cli DRC JSON output stores coordinates in items[].pos.x/y,
not at the violation top level. The code was reading violation.x/y
which don't exist, falling back to (0, 0) for every violation.
Now correctly reads the position from the first item in each
violation entry.
Two issues fixed:
1. TypeScript schema was missing the outline parameter entirely,
so MCP clients couldn't send pour boundary points.
2. Python code read "points" key but schema defined "outline" key.
Now accepts "outline" (with "points" as fallback for backwards
compatibility). When no outline is provided, automatically uses
the board edge bounding box as the pour boundary.
NETINFO_ITEM objects don't have a GetClassName() method, causing
an AttributeError crash when listing nets. The correct method is
GetNetClassName() which returns the net class name string.
str(module.GetFPID()) returns the Swig proxy representation like
"<pcbnew.LIB_ID; proxy of ...>" instead of the actual footprint name.
GetUniStringLibId() returns the proper "Library:Footprint" string.
Convenience wrapper around route_trace that eliminates the need for
separate get_pad_position calls before routing.
- Accepts fromRef/fromPad/toRef/toPad instead of raw coordinates
- Automatically looks up pad positions from board footprints
- Auto-detects net from pad assignment (overridable via net param)
- Returns fromPad/toPad position info in response
- Saves ~2 tool calls (64+ calls for a full TMC2209 board) vs 3-step flow
Registered in: routing.py, kicad_interface.py (dispatch), routing.ts (MCP)
- fix: DynamicSymbolLoader reads project sym-lib-table before global dirs
add_schematic_component now finds symbols from project-local .kicad_sym files
project_path derived automatically from schematic file path
- fix: place_component reloads FootprintLibraryManager with project_path
new boardPath parameter passed to place_component tool (TypeScript + Python)
_handle_place_component wrapper recreates LibraryManager per project
- fix: copy_routing_pattern geometric fallback when pads have no nets
primary filter: net-based (when pads are assigned to nets)
fallback: bounding box of source footprint pads +5mm tolerance
filterMethod field in response indicates which mode was used
- feat: register copy_routing_pattern as MCP tool in routing.ts
sourceRefs, targetRefs, includeVias, traceWidth parameters
Live tested: ESP32 + 2x TMC2209 in Test3 project
13 traces U2 routed, copy_routing_pattern copied all 13 to U3
offset Y+30mm correct, 26 total traces verified
- Update schematic version from 20230121 (KiCAD 7) to 20240101 (KiCAD 9)
- Replace invalid all-zeros UUIDs in template files with valid UUIDs
- Add post-copy UUID regeneration to ensure each project gets unique UUID
- Add UTF-8 encoding and Unix line endings for cross-platform compatibility
Fixes the root cause where templates were copied with invalid UUIDs and
outdated version format, causing KiCAD 9.0.x to reject the schematic files.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Bug 1 - add_schematic_component: footprint parameter silently ignored
The footprint value from MCP params was never passed through to
DynamicSymbolLoader.add_component() / create_component_instance().
Every placed symbol had an empty Footprint field regardless of input.
Fix: added footprint: str='' to both functions, passed through all
call sites, added footprint to schematic.ts tool schema.
Bug 2 - delete_schematic_component: only deleted first duplicate
When a reference appeared multiple times (e.g. after a failed
add attempt), only the first instance was removed due to break
after first match. Fix: collect all matching blocks first, then
delete back-to-front to preserve indices. Response now includes
deleted_count.
New tool - edit_schematic_component
Update footprint, value or reference of a placed symbol in-place.
More efficient than delete+re-add: preserves position and UUID.
Accepts: schematicPath, reference, footprint?, value?, newReference?
All 3 fixes verified by live tests on a real JLCPCB/KiCAD 9 project:
- R_TEST1: footprint Resistor_SMD:R_0603_1608Metric written correctly
- J1 duplicate: deleted_count=2 with single call
- J2 edit: PinSocket footprint assigned in-place, no delete+add needed
- PCB update (F8) confirmed: only components with footprint imported
Verified by live test: created a fresh schematic via MCP, result shows
format version 20250114, 0 _TEMPLATE_ hits, 0 (lib_id -100) hits, 24
real components placed cleanly.
--- Format version bump (20230121 KiCAD 7 -> 20250114 KiCAD 9) ---
Files: python/templates/*.kicad_sch, python/commands/project.py,
python/commands/schematic.py
Root cause: the MCP server targets KiCAD 9 exclusively - pcbnew.pyd is
compiled for KiCAD 9.0 / Python 3.11.5, and server.ts explicitly selects
the KiCAD 9 bundled Python on Windows. Generating new schematics with a
2-year-old format tag caused a spurious 'This file was created with an
older KiCAD version' warning on every newly created schematic.
--- Remove corrupt _TEMPLATE_* placed-symbol blocks ---
File: python/templates/template_with_symbols_expanded.kicad_sch
Root cause: the expanded template was generated by the old sexpdata
serializer (same corruption PR #40 fixed for DynamicSymbolLoader add-path).
The serializer converted the string 'Device:R' to integer -100, producing
(lib_id -100) instead of (lib_id Device:R). KiCAD cannot resolve an
integer as a library reference and crashes with a null-pointer when the
user attempts to select these symbols. They appeared as grey _TEMPLATE_R?,
_TEMPLATE_C?, _TEMPLATE_U_REG? etc. ~5000mm off-sheet - invisible during
normal work but triggering a crash on accidental box-select.
Discovered via live testing on a real JLCPCB/KiCAD 9 project.
