Local-first, AI-extensible PCB automation through the Model Context Protocol
CopperMCP exhibits significant definition quality gaps. While all 30 tools have names and descriptions, the descriptions are highly technical and domain-specific (PCB design jargon: 'Board IR', 'ERC', 'DRC', 'route-bundle'), making them difficult for general LLMs to understand when to invoke them. Input schemas are present but minimal, most parameters are typed as generic 'object' with only brief descriptions like 'Structured circuit content' or 'Optional Settings configuration', providing no guidance on internal structure, required fields, or valid values. No tools declare output schemas. Error handling is absent from descriptions. Parameter descriptions lack actionable constraints (ranges, enums, formats). The server prioritizes domain correctness over LLM usability.
Operator-gated, token-authorized route-candidate apply with atomic replacement.
Operator-gated, token-authorized live KiCad IPC route apply.
Operator-gated, token-authorized bounded placement apply with atomic replacement.
Cancel an in-progress routing job.
Deterministically rank and compare multiple candidates.
Authorization-bound candidate geometry export.
Get the status and results of a routing job.
Generic object-typed parameters with minimal descriptions. 28 of 30 tools accept 'payload', 'content', or 'settings' parameters typed as 'object' with descriptions like 'Structured circuit content' or 'Optional Settings configuration'. No documentation of required fields, nested structure, or valid values. LLMs cannot construct valid payloads without trial-and-error.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Inspect a KiCad board beneath the configured workspace.
Describe one board's Board IR structure without disclosing its content.
Observe one open KiCad PCB through the optional local IPC adapter.
Return a detached revision-bound live editor context.
Inspect active layer and native selection refs without board text or mutation.
Observe a region-scoped semantic Circuit Scene with quarantined board text.
Observe a region-scoped semantic Circuit Scene from live KiCad IPC.
Revision-bound post-placement scene and DRC observation (read-only).
Revision-bound layered route proposal (read-only).
Revision-bound live layered route proposal (read-only).
Revision-bound live KiCad IPC placement proposal (read-only).
Revision-bound live KiCad IPC route proposal (read-only).
Typed placement intent with deterministic legality preview.
Non-mutating two-pin route preview on a documented Board IR subset.
Revision-bound atomic multi-net route-bundle preview (read-only).
Validate structured circuit content and return a private schematic build.
Run authoritative KiCad DRC for a board beneath the configured workspace.
Describe implemented and planned capabilities without overstating maturity.
Start a durable file-backed layered routing job with bounded worker execution.
Validate a candidate manifest against the schema.
Render structured circuit content and check it with the authoritative KiCad ERC.
Verify one versioned external route document with mandatory read-only KiCad DRC.
Check whether a workspace board implements structured circuit content's connectivity.
No output schemas documented. Tool descriptions do not specify what fields or structure the response contains. LLMs cannot plan downstream tool calls or extract required data (e.g., job_id from start_routing_job) without guessing.
Domain-specific jargon in descriptions without explanation. Terms like 'Board IR', 'ERC', 'DRC', 'route-bundle', 'revision-bound', 'quarantined board text', 'Circuit Scene' are unexplained. General LLMs unfamiliar with PCB design cannot determine when to invoke these tools or what they accomplish.
No error handling guidance in descriptions. Tools do not document what errors can occur, when to retry, or what the LLM should do if a call fails. E.g., 'apply_candidate' is 'Operator-gated, token-authorized' but does not explain what happens if authorization fails or how to recover.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). 23 tools are READ_ONLY and 5 are WRITE, but this is not declared in the tool definitions. LLMs cannot infer safety properties and may incorrectly retry destructive operations or assume read-only tools are safe to call repeatedly.