Setup Guide

Get your AI agent connected to VisoWork's vision tools. Choose the method that fits your workflow — install the skill package directly, connect via MCP, or both.

Setup Skill

Pick a skill bundle and tell your agent to install it. Start with the starter for the core vision tools, then add domain skills from the catalog as your use case grows.

Starter: VisoWork Vision Skill

VisoWork Vision Skill

~6 KB

Three vision tools — describe, OCR, and object detection — wrapping the VisoWork REST API.

vision.describevision.ocrvision.detect

/skills/visowork-toolkit.zip

Download .zip

Ask your AI agent to install the downloaded package:

bash

# Install directly from the URL:
> Install the VisoWork vision skill from
  https://visowork.com/skills/visowork-toolkit.zip

# Or from a local file after downloading:
> Install the skill from ./visowork-toolkit.zip

Once installed, your agent has access to all vision analysis tools through natural conversation. The starter wraps three tools over the VisoWork REST API:

Describe — Natural-language descriptions of images
OCR — Text extraction with block-level positions
Detect — Object localization with bounding boxes and confidence

Domain Skills from the Catalog

Need something beyond the starter — retail shelf audits, manufacturing QA, medical document parsing? Browse /skills and filter by domain to find the right bundle.

Every catalog skill has a stable download URL in the same pattern:

text

https://visowork.com/skills/<skill-name>.zip
# or pinned to a specific release:
https://visowork.com/skills/<skill-name>-<version>.zip

Open any skill's detail page for a one-click Download button (or copy the URL directly), then install the same way as the starter:

bash

# Install from the catalog URL:
> Install the skill from
  https://visowork.com/skills/shelf-audit.zip

# Pin to a specific version:
> Install the skill from
  https://visowork.com/skills/shelf-audit-1.0.0.zip

# Or from a local file after downloading:
> Install the skill from ./shelf-audit.zip

Domain skills are self-contained bundles — the install flow is identical, and your agent picks the right tool based on the request. Mix and match as many as you need; skills coexist cleanly side-by-side.

Setup MCP

Connect via MCP protocol for direct tool access from Claude Code, Cursor, or other MCP clients.

MCP setup is optional. If you installed the skill, your agent already has access to all vision tools.

Claude Code

Add VisoWork as an MCP server in Claude Code:

bash

# Add VisoWork as an MCP server in Claude Code
claude mcp add visowork \
  --transport sse \
  --url https://api.visowork.com/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

# Verify it's connected
claude mcp list

Other MCP Clients

Add the following to your MCP client configuration file (Claude Desktop, Cursor, Windsurf, etc.):

json

{
  "mcpServers": {
    "visowork": {
      "url": "https://api.visowork.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

MCP Endpoint

Endpoint: https://api.visowork.com/mcp

All MCP requests require a valid API key in the Authorization header. Generate one in your API Keys settings.

Use on Agent Chat

Once the skill is installed or MCP is connected, set your API credentials and start analyzing.

Environment Setup

bash

# Set your API endpoint and key before starting the agent
export VISOWORK_API_URL=https://api.visowork.com
export VISOWORK_API_KEY=YOUR_API_KEY

Agent Chat Examples

bash

# Then use natural language in your agent:
> Describe this image: /path/to/photo.jpg
> Extract text from this receipt image
> Detect all objects in this screenshot
> What objects are in this photo? Focus on vehicles.

Your agent will automatically select the best vision tool based on your natural language request.

Agent Usage

If you're authoring a skill or integrating another tool, these three conventions keep things predictable across VisoWork:

Use project-scoped MCP for platform capabilities. Storage, vector search, and vision are all reachable via your project's MCP config — configure once in your MCP client and keep the surface consistent across skills.
Start from the skill template. Copy sandbox/skills/template/ (or use the AI skill creator in the admin dashboard, which extends it for you). Your skill's scripts/cli.py is its capability set; declare Python dependencies inline via PEP 723.
Precheck before running. The template's precheck() validates VISOWORK_API_KEY and the script's own presence, and prints a single-line JSON hint to stderr ({"error":"precheck_failed","hint":"..."}) on failure. Add your skill-specific checks inside precheck().

A fully-worked reference skill lives at sandbox/skills/builtin/visowork-toolkit/.

Need an API Key?

Generate and manage API keys for your projects in the API Keys settings.

Overview Catalog