Getting Started with Caliper via MCP


Caliper is a geometry file metadata API that your AI agent can call as an MCP tool. Send a 3D file, get structured metadata back — bounding boxes, triangle counts, surface area, volume, manifold analysis, and more.

This guide walks you through connecting to Caliper via MCP and making your first analysis call.

What you’ll need

  • An MCP client (Claude Desktop, Cursor, Windsurf, or any MCP-compatible agent framework)
  • A USDC wallet on Base for payments (format detection is free)

Step 1: Add Caliper to your MCP client

Caliper’s MCP endpoint is:

https://caliper.fit/mcp/

Claude Desktop

Add this to your Claude Desktop MCP configuration (claude_desktop_config.json):

{
  "mcpServers": {
    "caliper": {
      "url": "https://caliper.fit/mcp/"
    }
  }
}

Restart Claude Desktop. Caliper’s 9 tools will appear in your tool list.

Cursor / Windsurf

Add the MCP server URL https://caliper.fit/mcp/ through your editor’s MCP server settings.

Programmatic MCP clients

If you’re building an agent with an MCP client library, point it at https://caliper.fit/mcp/. The manifest at https://caliper.fit/mcp/manifest describes all available tools.

Step 2: Try format detection (free)

The caliper_detect_format tool is free — no payment setup needed. Use it to verify your connection works.

Ask your agent:

“Use the caliper_detect_format tool to identify this file: https://raw.githubusercontent.com/mikedh/trimesh/main/models/featuretype.STL

You should get back a response identifying the file as STL format with details about the file.

Step 3: Set up x402 payment

Most Caliper tools require payment via the x402 protocol. When you call a paid tool without payment, the server responds with HTTP 402 and payment instructions.

How x402 works:

  1. Your MCP client calls a Caliper tool
  2. If payment is required, the server returns a 402 response with a payment address and amount
  3. Your client pays the specified amount in USDC on Base
  4. The request is retried with a payment proof header
  5. The server verifies payment and returns results

MCP clients with built-in x402 support handle this automatically. If your client doesn’t support x402 natively, you can use the REST API with manual payment (see the REST Getting Started guide).

Step 4: Analyze your first file

Once payment is set up, try the recommended entry point — caliper_auto_stats:

“Use caliper_auto_stats to analyze this STL file: https://raw.githubusercontent.com/mikedh/trimesh/main/models/featuretype.STL

Caliper will auto-detect the format and return structured metadata:

{
  "format": "stl_binary",
  "vertices": 1206,
  "triangles": 2400,
  "bounding_box": {
    "min": [-5.0, -5.0, 0.0],
    "max": [5.0, 5.0, 10.0]
  },
  "surface_area": 478.32,
  "volume": 392.67,
  "is_manifold": true,
  "is_watertight": true
}

(Actual values will vary based on the file analyzed.)

Step 5: Batch processing

Need to analyze multiple files? Use caliper_batch_stats to process up to 10 files in a single request:

“Use caliper_batch_stats to analyze these files:

This is more efficient than individual calls — one payment covers the entire batch.

Available tools

ToolWhat it doesCost
caliper_detect_formatIdentify file formatFree
caliper_auto_statsAuto-detect format, extract full metadataPaid
caliper_stl_statsSTL-specific analysisPaid
caliper_obj_statsOBJ file analysisPaid
caliper_ply_statsPLY file analysisPaid
caliper_pcd_statsPCD point cloud analysisPaid
caliper_las_statsLAS/LAZ point cloud analysisPaid
caliper_gltf_statsglTF/GLB analysisPaid
caliper_batch_statsBatch process up to 10 filesPaid

Input methods

You can send files to Caliper two ways:

  • File URL — pass an HTTP/HTTPS URL pointing to the file (up to 100 MB)
  • Base64 — pass the file content as a base64-encoded string (up to 200 KB decoded)

URLs are recommended for larger files. Base64 is convenient when you already have the file content in memory.

What’s in the metadata?

The exact fields depend on the file format, but typically include:

  • Geometry bounds — bounding box (min/max coordinates)
  • Mesh statistics — vertex count, face/triangle count, edge count
  • Surface properties — surface area, volume (for watertight meshes)
  • Topology — manifold status, watertight check
  • Point cloud stats — point count, density, classification distribution (for PCD/LAS)
  • Scene structure — node hierarchy, material count, animation count (for glTF)

Next steps