.svg){kind=small}
How to Give Your AI Agent Image Tools (Claude, Cursor, ChatGPT) — 2026
Your agent can already reason about images. To let it actually process them — compress, convert, describe, OCR, turn into a PDF — connect an image MCP server. Here is how, for Claude, Cursor and beyond, in a few minutes.
Table of Contents
Quick answer
To give an AI agent image tools, connect a remote MCP server. In an MCP client like Claude or Cursor, add the server URL http://omnikitbolt.com/api/mcp, approve the OAuth sign-in (no API key to paste), and the agent discovers 20+ image and PDF tools plus AI vision automatically. For clients without MCP support, call the same tools through the REST API with a generated key.
What you need
An MCP-compatible agent client (Claude desktop or web, Cursor, or similar) and a Google account for the one-time sign-in. That is it — there is no SDK to install and no key to manage for the MCP path.
Claude Desktop / Cursor (config file)
Add the server to your client's MCP configuration:
{
"mcpServers": {
"omnikitbolt": { "url": "http://omnikitbolt.com/api/mcp" }
}
}Restart the client. The first time the agent uses a tool, it opens the OAuth flow: sign in with Google, approve, and you are connected. If you prefer a key instead of OAuth, add an Authorization header:
{
"mcpServers": {
"omnikitbolt": {
"url": "http://omnikitbolt.com/api/mcp",
"headers": { "Authorization": "Bearer sk_live_YOUR_KEY" }
}
}
}Claude on the web (connector)
In Claude's settings, open Connectors → Add custom connector. Give it a name (omnikitbolt) and paste the MCP server URL http://omnikitbolt.com/api/mcp. Continue, sign in with Google when prompted, and the tools appear. Nothing else to configure.
Other agents: the REST API
If your framework does not support remote MCP yet, call the same tools over plain HTTP. Generate a key on the API dashboard and hit an endpoint:
curl -X POST http://omnikitbolt.com/api/v1/convert \
-H "Authorization: Bearer sk_live_YOUR_KEY" \
-F "file=@photo.jpg" -F "format=webp" -F "quality=80" \
-o out.webpFirst prompts to try
- "Compress this image and convert it to WebP."
- "Describe this image and generate alt text and an SEO filename."
- "Extract the text from this screenshot." (OCR)
- "Optimize this photo for the web." (one-call pipeline)
- "Turn these images into a single PDF."
For chaining several edits in one call, see how to build image pipelines.
How billing works
Every account has a free daily tier. Beyond that it is pay-as-you-go with prepaid credits: most operations cost 1 credit, AI vision ops (describe, alt text, OCR, tags, filename) cost 2, and a pipeline is 1 per step. A failed operation is refunded automatically, and credits are shared across the website and the API/MCP. To compare the economics with other services, see our image API comparison for agents.
