Skip to content

MCP.md

Latest commit

 

History

History
75 lines (60 loc) · 3.2 KB

MCP.md

File metadata and controls

75 lines (60 loc) · 3.2 KB
Raw

WireViz MCP server

The WireViz engine is exposed as a Model Context Protocol server so any agent can build and verify wire harnesses — design a harness in YAML, validate it, run design-rule checks, compute the BOM / cut sheet / netlist / weight, recommend a wire gauge, and render a diagram — iterating on structured results rather than scraping CLI output.

Install & run

pip install "wireviz[mcp]"     # adds the optional `mcp` dependency
wireviz-mcp                     # runs the server over stdio
# or: python -m wireviz.wv_mcp

Register it with an agent

Point any MCP client at the wireviz-mcp command. For a Claude Code / Claude Desktop style config (.mcp.json or claude_desktop_config.json):

{
  "mcpServers": {
    "wireviz": {
      "command": "wireviz-mcp",
      "env": { "WIREVIZ_MCP_OUTPUT_DIR": "/path/to/harness-output" }
    }
  }
}

WIREVIZ_MCP_OUTPUT_DIR is optional (see Where the output lives below).

Tools

Tool Purpose
validate_harness(harness_yaml) parse + summary (or a parse error) — call first
run_drc(harness_yaml) design-rule findings ranked by severity
generate_bom(harness_yaml) bill of materials rows
generate_cutsheet(harness_yaml) per-wire cut list + bulk length by gauge
generate_netlist(harness_yaml) electrical nets + floating pins
render_svg(harness_yaml) native grid-snapped SVG (inline text)
render_diagram(harness_yaml, fmt, output_dir) Graphviz svg/png/html (needs dot)
recommend_gauge(current, length_m, max_drop_v) minimum AWG for a run
engineering_report(harness_yaml) weight, bundle/sleeve, power/voltage-drop
list_connectors() / list_devices() built-in libraries
import_wirelist(csv_text) / import_kicad(netlist_text) seed a harness from a wire list / netlist → YAML
generate_formboard(harness_yaml, page, output_dir) 1:1 formboard SVG

Every tool is stateless — it takes the full harness YAML each call (harnesses are small text) — and returns { "ok": true, ... } or { "ok": false, "error": ... }.

Where the output lives

By default, nowhere on disk — output is returned inline in the tool result. render_svg and generate_formboard return the SVG as text; generate_bom, run_drc, generate_netlist, etc. return JSON. The calling agent receives the data and decides where to save it, so the server works even when the agent has no shared filesystem (sandboxed or remote).

The file-oriented tools (render_diagram for png/html, and optionally generate_formboard) take an output_dir argument — or read the WIREVIZ_MCP_OUTPUT_DIR environment variable — and when set they write the file there and return its path instead of the content. So: inline by default, on-disk when you ask for it.

A typical agent design loop

  1. validate_harness — confirm the YAML parses.
  2. run_drc — fix any errors/warnings in the findings.
  3. recommend_gauge — size power wires for their current/length.
  4. generate_bom / generate_cutsheet / engineering_report — costing & build data.
  5. render_svg (or render_diagram) — a diagram to show the human.