Give your AI agent CAD vision.

The hosted server speaks Streamable HTTP and authenticates with the same cadl_ key you already use for REST. Add it to Claude Code:

claude mcp add --transport http cadlens \
  https://api.cadlens.co/mcp \
  --header "Authorization: Bearer cadl_your_key_here"

Or drop it into Claude Desktop's config, running locally over stdio:

{
  "mcpServers": {
    "cadlens": {
      "command": "npx",
      "args": ["-y", "cadlens-mcp"],
      "env": { "CADLENS_API_KEY": "cadl_your_key_here" }
    }
  }
}

Then just ask: “Parse this floor plan and tell me how many circles are on the ELECTRICAL layer.”

Each one wraps the same REST endpoint documented in our API reference, with identical authentication, ownership checks, and plan quotas.

• parse_cad_file — Upload a DWG, DXF, or DWF file and parse it. Waits for the result by default.
• get_job_status — Check whether a parse is pending, processing, complete, or failed.
• get_job_result — Read the vector JSON — summary first, then drill into a layer or entity type.
• get_job_image — Get pre-signed PNG preview URLs, one per sheet.
• list_jobs — List recent jobs on the account, newest first.
• delete_job — Delete a job and its artifacts.
• get_usage — Plan, quota, requests used, and remaining for the current period.

A single drawing can carry tens of thousands of entities. Dumping that into a model is expensive and usually useless, so get_job_result is progressive by default:

• summary (default) — file info, per-sheet and per-layer counts, layer names and colours, bounding boxes. No geometry.
• entities_on_layer / entities_by_type — a filtered slice, capped at 500 entities, with an honest truncated flag when there is more.
• full — everything, when the agent really does need everything.

The agent looks before it leaps: read the summary, pick the layer that matters, pull only those entities.

An agent that has never heard of CADLens can find its way in without a human. Everything below is public, unauthenticated, and cacheable:

• /.well-known/mcp/server-card.json — both transports, the auth scheme, every tool.
• /.well-known/agent-skills/index.json — SKILL.md manifests with SHA-256 digests.
• /.well-known/api-catalog — RFC 9727 linkset to docs, OpenAPI, and Postman.
• /llms.txt and /llms-full.txt — plain-text context, no markup to strip.
• Accept: text/markdown — pages return markdown instead of HTML.
• WebMCP — browser agents get in-page tools on navigator.modelContext to search these docs and navigate the site.