Skip to content
Developer API

The MCP server and tools

The AgentHere MCP server: vision, OCR, audio, image/video generation, web search, and page-reader tools — and how to wire them into Claude Desktop and Cursor.

3 min readUpdated 2026-07-07

Besides Chat Completions, AgentHere provides an MCP server (Model Context Protocol) with a set of tools: image and video analysis, OCR and audio transcription, image and video generation, web search, and page reading. You can wire it into Claude Desktop, Cursor, and any MCP-capable client.

Endpoint

  • URL: https://agenthere.ru/api/mcp
  • Transport: Streamable HTTP (POST) and SSE (GET).
  • Auth: header Authorization: Bearer ah_YOUR_KEY (proxy or config key).

You can get the URL programmatically via /api/developer/info.

Tools

The server exposes eight tools across six categories: vision, audio, image generation, video generation, web search, and page reading.

Understanding (media → text)

  • analyze_image (model glm-5v-turbo) — analyze one or more images and answer a question about them: descriptions, UI/code extraction, diagrams, charts, screenshots, and text from documents/screenshots;
  • analyze_video (model glm-4.6v) — analyze a video and answer questions about its content, scenes, and events;
  • transcribe_audio (model glm-asr-2512) — transcribe speech in an audio file to text (WAV/MP3, up to 25 MB).

Generation (text → media)

  • glm_generate_image (model glm-image) — generate an image from a text prompt and return a stable image URL;
  • glm_generate_video (model cogvideox-3) — start generating a short video from a text prompt and/or a reference image; returns a task_id immediately;
  • glm_get_generation — poll a glm_generate_video task until the video URL is ready (free).

Web

  • web_search (model search-prime) — search the web and return structured results (titles, URLs, snippets);
  • web_reader — fetch and read a web page, returning its content as markdown or text.

Wiring it into Claude Desktop

Add the server to the mcp.json config file:

json
{
  "mcpServers": {
    "agenthere": {
      "url": "https://agenthere.ru/api/mcp",
      "headers": {
        "Authorization": "Bearer ah_YOUR_KEY"
      }
    }
  }
}

After a restart, AgentHere's tools appear in the available list — the model can call them automatically.

Wiring it into Cursor

In MCP settings add a new server with the same URL and auth header. Cursor will offer the vision, OCR, generation, and search tools as you work with code, screenshots, and documents.

Metering

Every tool call checks the balance and accounts for token usage and latency — just like chat. With a zero balance the call is rejected. More in Billing & tokens.

Uploading media

For tools that take an image, video, or audio input, upload the file via POST /api/mcp/media/upload (a proxy or config key) — the endpoint returns a media reference ({kind}/{filename}) or a public URL that you pass to the tool. Image analysis also accepts a public HTTPS URL directly.

The MCP server and tools | AgentHere