Skip to main content
Glama

pictostl-mcp

Local stdio MCP server for pictostl.com. Generation runs on pictostl.com — this package uploads photos, starts 3D jobs, and writes GLB or STL files on the machine that hosts the MCP server.

Results are AI-reconstructed meshes for visualization and 3D printing experiments. They are not engineering drawings.

This is a local stdio MCP server distributed from its GitHub repository. pictostl.com does not host a remote MCP HTTP endpoint.

Architecture and data flow

MCP client → local pictostl-mcp process → pictostl.com API
                                      ← generation status and GLB
MCP client ← local GLB/STL file       ←

The MCP process runs locally and communicates with https://pictostl.com over HTTPS. Images, generation parameters, and the API key are sent to pictostl.com to upload inputs, create jobs, and retrieve results. GLB-to-STL conversion and final file writing happen locally on the machine running the MCP server.

There is no hosted PicToSTL MCP endpoint. MCP clients start this package locally over stdio; the local process then calls the hosted PicToSTL API.

Related MCP server: Meshy MCP Server

Get an API key

  1. Sign up

  2. Create a key at API keys

  3. Buy credits on Pricing

By using the API or this MCP server you agree to the Terms of Service.

Website browser generation is separate from MCP. MCP always uses an API key.

Install from GitHub

Requires Node.js 20 or newer. The command below installs the repository through npm's Git support and runs its committed JavaScript build. It does not require a published npm package or a local TypeScript compiler.

Claude / Cursor config

{
  "mcpServers": {
    "pictostl": {
      "command": "npx",
      "args": [
        "-y",
        "--package=github:qduoduo-hwh/pictostl-mcp",
        "pictostl-mcp"
      ],
      "env": {
        "PICTOSTL_API_KEY": "ps_live_..."
      }
    }
  }
}

Set PICTOSTL_API_KEY to a live key from https://pictostl.com/settings/api-keys.

Optional: PICTOSTL_API_BASE (default https://pictostl.com) for a local website.

Keep the API key in your MCP client's secret or user-level configuration. Do not commit a live key to source control.

For a pinned installation, append a Git tag or commit after the repository name, for example github:qduoduo-hwh/pictostl-mcp#v0.1.0.

Local development

git clone https://github.com/qduoduo-hwh/pictostl-mcp.git
cd pictostl-mcp
corepack enable
pnpm install --frozen-lockfile
pnpm build

Then configure the MCP client to run node with the absolute path to dist/index.js as its first argument.

Tools

Seven MCP tools:

Tool

Purpose

list_generation_options

Quality, texture, multi-view rules, and credit costs

upload_image

Upload a local file or HTTPS URL; returns assetId

generate_model

Start an async 3D job (sends clientRequestId even if you omit it)

get_task

Fetch a generation by task id; optional waitSeconds (max 60)

list_my_generations

List recent generations for this API key

get_account

Remaining credits and Pro / Ultra paid-credit status

download_model

Write original GLB, or a local binary STL, to a path

Typical flow: upload_image → generate_model → get_task (repeat with the same task id if it takes more than 60 seconds) → download_model with that task id.

Do not pass an upload assetId to get_task or download_model.

download_model default is format=stl, longest edge 100 mm, converted locally from the original GLB (not the website preview mesh). STL has geometry only — no color or textures. format=glb writes the original file unchanged.

Reuse clientRequestId from a previous generate_model result when retrying the same job.

Contributing

Improve tool descriptions so agents call the right tool with the right fields. Do not ask for internal fal prompts or provider payloads.

License

MIT

Available Tools

7 tools
download_modelA

Download a completed generation to a local path. id is the task id from generate_model/get_task, not an upload assetId. format stl (default) converts the original GLB locally to binary STL with longest edge 100 mm (no color/texture). format glb writes the original file. Parent directory must already exist.

Inputs: id (required), path (required), format, longestEdgeMm.

generate_modelA

Start an async image-to-3D job on pictostl.com. Requires imageAssetIds from upload_image. If you omit clientRequestId the server call still sends one — reuse that value to retry the same job without a second charge. Poll with get_task using the returned task id.

Inputs: mode (required), quality (required), texture (required), imageAssetIds (required), +2 more.

get_accountA

Show remaining credits and whether Pro / Ultra can be paid with non-trial credits.

Inputs: no named parameters.

get_taskA

Fetch a generation by task id (not an upload assetId). Optional waitSeconds (max 60) polls until completed or failed. Jobs often take longer than 60 seconds — call again with the same id until it is completed or failed.

Inputs: id (required), waitSeconds.

list_generation_optionsA

List pictostl.com generation options available to API clients: quality (Basic/Pro/Ultra), texture, single vs multi-view rules, and credit costs.

Inputs: no named parameters.

list_my_generationsB

List recent 3D generations for this API key's user.

Inputs: no named parameters.

upload_imageA

Upload a local image file or HTTPS URL and return an assetId for generate_model. Do not pass this id to download_model or get_task.

Inputs: url, path.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.2/5.0

    Scored across 7 tools

    Disambiguation5/5

    Each tool maps to a distinct step in the image-to-3D pipeline: options, upload, generate, list, poll, account, download. Even get_task and list_my_generations differ clearly (single fetch by ID vs. list history). No two tools have overlapping purposes.

    Naming Consistency5/5

    All tool names follow a consistent snake_case verb_noun pattern (list_, upload_, generate_, get_, download_). The convention is uniform and predictable, making the API easy to navigate.

    Tool Count5/5

    Seven tools cover the full lifecycle of the service without excess. Each tool serves a necessary function in the workflow, and the count is well within the ideal range.

    Completeness5/5

    The tool surface covers the complete flow: discover options, upload, generate, poll, download, list history, and check account. No critical operations are missing for an image-to-3D generation API.

    Maintenance

    ActivitySlowing
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers