zai-tools-lite

Lean Z.ai tools for the pi coding agent: URL reader, GitHub repo lookup (zread), and image/video vision. Minimal token footprint.

Intentionally no search tool — pair with pi-web-access for that.

Tools

z_ai_reader

Fetch a URL through Z.ai's web reader MCP and get Markdown or plain text back.

| Param | Type | Notes | |---|---|---| | url | string | URL to fetch | | return_format | "markdown" | "text" | optional | | no_cache | boolean | optional | | retain_images | boolean | optional | | timeout | integer | optional |

z_ai_zread

Public GitHub repo Q&A and structure lookup via Z.ai's zread MCP.

| action | Required params | Optional | |---|---|---| | search_doc | repo_name, query | language (en/zh) | | read_file | repo_name, file_path | | | get_repo_structure | repo_name | dir_path |

z_ai_vision

Analyze an image or video (MP4/MOV/M4V, ≤8MB) through Z.ai Vision. Pick the action matching the input:

• analyze_image, extract_text_from_screenshot, diagnose_error_screenshot, understand_technical_diagram, analyze_data_visualization — need image_source (path or URL) + prompt
• analyze_video — needs video_source + prompt
• ui_to_artifact — needs image_source, output_type (code/prompt/spec/description)
• ui_diff_check — needs expected_image_source + actual_image_source

Optional extras: programming_language, context, diagram_type, analysis_focus.

Install

pi install npm:zai-tools-lite

Also on npm: zai-tools-lite

Authentication

Provide a Z.ai API key either way:

• ZAI_API_KEY (or Z_AI_API_KEY) environment variable, or
• log in to the zai provider in pi (/login → zai)

The vision tool spawns the official @z_ai/mcp-server over stdio; reader and zread talk to Z.ai's hosted MCP endpoints over HTTP.

Configuration

• Z_AI_MCP_TIMEOUT_MS — per-call timeout in milliseconds (default: 180000).

License

MIT