matrix MCP Server 开发者文档

Synthesize images from text prompts, with optional reference images for editing/transformation. Max 10 requests per call. REQUIRED format — wrap each request in a 'requests' array (do NOT pass prompt as flat parameter): {"requests": [{"prompt": "A cute cat", "output_file": "cat.png", "aspect_ratio": "1:1"}]} For image editing, pass reference images: {"requests": [{"prompt": "Make it watercolor style", "input_urls": ["https://..."], "output_file": "out.png"}]}

// 输出示例(content[0].text)
{"output_files":["cat.png"],"status":"ok"}

Analyze images with prompts and return text results. Max 20 images. Supports: jpeg, png, webp. REQUIRED format — use 'image_info' array (NOT 'images', 'image_list', or 'image_urls'): {"image_info": [{"url": "https://...", "prompt": "Describe this image"}]} {"image_info": [{"file": "/absolute/path/to/img.png", "prompt": "OCR this"}]}

// 输出示例(content[0].text)
{"results":[{"description":"mock description","file":"i.png"}]}

Search and download images. At most 10 queries per request. This method is designed to be robust - individual task failures will not affect other tasks. Safety Check: - All downloaded images are automatically checked for content safety. - Images that fail safety check will be deleted and excluded from results. - DO NOT retry downloading images that failed safety check.

// 输出示例(content[0].text)
{"results":[{"files":["mock_image_1.png"],"query":"cat","task_name":"cat"}]}

List available images in workspace. Returns a list of image file paths.

// 输出示例(content[0].text)
# Total Images: 11

## Image profile.png
File: profile.png
Path: /workspace/profile.png
File Size: 1.0 KB

## Image og.png
File: og.png
Path: /workspace/og.png
File Size: 1.0 KB

## Image banner.png
File: banner.png
Path: /workspace/banner.png
File Size: 1.0 KB

## Image thumb.png
File: thumb.png
Path: /workspace/thumb.png
File Size: 1.0 KB

## Image hero.png
File: hero.png
Path: /workspace/hero.p
…

Reverse image search. Returns similar images and related search queries.

// 输出示例(content[0].text)
{"image_url":"https://example.com/i.png","output_file":"out.md","status":"ok"}

Generate video files from text prompts or images. USE THIS TOOL when user asks to "create a video", "make a clip", "generate video". Unified entry point for text-to-video (T2V) and image-to-video (I2V). Max 5 requests per call. Duration: 6 or 10 seconds. Resolution: 768P or 1080P. Output: downloadable video file (.mp4). REQUIRED format — use 'video_requests' array (NOT 'requests', 'videos', or flat parameters): Text-to-video: {"video_requests": [{"prompt": "A cat playing", "output_file": "cat.mp4"}]} Image-to-video: {"video_requests": [{"prompt": "Cat runs", "output_file": "run.mp4", "image_file": "cat.png"}]} Prefer this tool over batch_text_to_video / batch_image_to_video — it handles both T2V and I2V.

// 输出示例(content[0].text)
{"output_files":["cat.mp4"],"status":"ok"}

Batch generate videos from text prompts (parallel-list format). Max 5 per call. Duration: 6 (default) or 10 seconds. Resolution: 768P (default) or 1080P. Prefer gen_videos for simpler usage — this tool uses parallel lists that must all match 'count'. REQUIRED: count, prompt_list, output_file_list (all lengths must equal count). Example: {"count": 2, "prompt_list": ["Sunset", "Ocean"], "output_file_list": ["s.mp4", "o.mp4"]}

// 输出示例(content[0].text)
{"count":1,"output_files":["cat.mp4"],"status":"ok"}

Generate videos from images. Max 5 requests per call. Duration defaults to 6s, resolution to 768P.

// 输出示例(content[0].text)
{"count":1,"output_files":["cat.mp4"],"status":"ok"}

Analyze and understand video content (describe, extract info, answer questions about video). NOT for generating video — use gen_videos or batch_text_to_video for generation. Maximum 10 videos per call. REQUIRED format — use 'video_info' array (NOT 'video_list', 'videos', or 'tasks'): {"video_info": [{"file": "/absolute/path/to/video.mp4", "prompt": "Describe this video"}]} {"video_info": [{"url": "https://cdn.example.com/video.mp4", "prompt": "Analyze"}]} File paths must be absolute (e.g. /Users/...), do NOT use file:// URIs.

// 输出示例(content[0].text)
{"results":[{"file":"v.mp4","prompt":"describe"}]}

Get all available voices for text-to-speech (TTS). Returns voice_id and voice_name. Call this before using synthesize_speech, batch_synthesize_speech, or batch_text_to_audio to choose a voice.

// 输出示例(content[0].text)
{"available_voices":[{"voice_id":"male-qn-qingse","voice_name":"青涩青年音色"},{"voice_id":"female-shaonv","voice_name":"少女音色"},{"voice_id":"female-yujie","voice_name":"御姐音色"}]}

Convert text to spoken audio (TTS / text-to-speech). USE THIS TOOL for voiceover, narration, dialogue, podcast, or reading text aloud. For music/song generation, use batch_text_to_music instead. Each parameter list should have the same order as text_list. Optional parameters default to None. Max 10 audios per call. Output: .mp3 audio files. Example: {"count": 2, "text_list": ["Hello world", "Welcome"], "output_file_list": ["hello.mp3", "welcome.mp3"]}

// 输出示例(content[0].text)
{"count":1,"output_files":["hello.mp3"],"status":"ok"}

Generate songs and music audio files from text prompts and lyrics. USE THIS TOOL when user asks to "create a song", "generate music", "make a beat", "compose", or any request involving music/song creation that expects an AUDIO FILE output (not just text lyrics). Lyrics are OPTIONAL — omit lyrics_list for pure instrumental/background music. Supports up to 1 minute per track, max 5 per call. Output: downloadable audio file (mp3/wav). When lyrics are provided, use newline ( ) to separate lines. Supports tags [Intro][Verse][Chorus][Bridge][Outro]. REQUIRED format — use parallel lists with matching 'count': With lyrics: {"count": 1, "prompt_list": ["Pop, sad"], "lyrics_list": ["[Verse] Rain falls..."], "output_file_list": ["song.mp3"]} Instrumental: {"count": 1, "prompt_list": ["Calm piano BGM"], "output_file_list": ["bgm.mp3"]}

// 输出示例(content[0].text)
{"count":1,"output_files":["bgm.mp3"],"status":"ok"}

Synthesize a single spoken audio clip from text (TTS). For music/song generation, use batch_text_to_music instead. Max text 10000 chars. Use get_voice_list for system voices or clone_voice for custom voices.

// 输出示例(content[0].text)
{"message":"Speech synthesis completed","output_file":"s.mp3","url":"https://mock-cdn.example.com/audio/s.mp3","url_clean":"https://mock-cdn.example.com/audio/clean_s.mp3","url_visible":"https://mock-cdn.example.com/audio/visible_s.mp3"}

Batch speech synthesis — generate multiple spoken audio clips at once (TTS). For music/song generation, use batch_text_to_music instead. Max 10 per call. Output files must be .mp3.

// 输出示例(content[0].text)
{"count":1,"output_files":["s.mp3"],"status":"ok"}

Analyze one audio deeply. Audio will not be displayed to the user. Maximum 1 audio file or URL per call. Supports: mp3, wav, aiff, aac, ogg, flac, m4a.

// 输出示例(content[0].text)
{"status":"ok","transcript":"mock transcript"}

Analyze and understand audio content (transcribe, describe, extract info from audio). NOT for generating audio — use batch_text_to_music or synthesize_speech for generation. Maximum 10 audios per call. REQUIRED format — use 'audio_info' array (NOT 'audios' or 'audio_list'): {"audio_info": [{"file": "/path/to/audio.mp3", "prompt": "Transcribe this"}]} {"audio_info": [{"url": "https://.../audio.mp3", "prompt": "Transcribe this"}]}

// 输出示例(content[0].text)
{"results":[{"file":"a.mp3","prompt":"transcribe"}]}

Extract content from multiple websites(mime type html) concurrently. The extracted content will be returned in JSON format. DO NOT use this tool to read local files! **NEVER** use this tool to extract pdf url, use `extract_pdfs_key_info` or `extract_pdfs_full_content` instead. Safety Check: - All extracted content will be checked for safety compliance. - If content fails safety check with "safety_check_failed" error, DO NOT retry or use alternative methods. - Safety check failures are final - abandon the extraction and move on to other tasks.

// 输出示例(content[0].text)
{"results":[{"content":"mock extracted content","task_name":"","url":"https://example.com/"}]}

Web Search API, works like Google Search. At most 10 queries per request. Search Strategy: - If valid results are fewer than 5, try searching page 2 by setting cursor=2 - If no valid results found, try changing search keywords - Valid results are those that pass safety checks (filtered results don't count) Safety Check: - All search results are automatically checked for safety compliance. - Results that fail safety check are automatically filtered out. - If all results are filtered, consider changing search keywords instead of retrying the same query.

// 输出示例(content[0].text)
{"results":[{"query":"hello world","results":[{"title":"Mock Result for hello world","url":"https://example.com/"}]}]}

Deploy the dist directory of Vite to web server. Note: - If no index.html is found, a warning is returned - Common development files (.git, node_modules, etc.) are automatically ignored - Returns a public URL where the deployed website can be accessed

// 输出示例(content[0].text)
{"website_id": 431264262663335, "website_url": "https://mock.example.com/diq164w5y1ju", "screenshot_url": ""}

Deploy a built website from an uploaded .tar.gz or .zip archive. Provide archive_url (server downloads it) or archive_data (base64-encoded archive). The archive should contain index.html at its root. Returns a public URL where the deployed website can be accessed.

// 输出示例(content[0].text)
{"website_id": 431166892779681, "website_url": "https://mock.example.com/8bgllq7v5j2x", "screenshot_url": ""}

Upload base64 file content to the CDN and return the public CDN URL. Use this to make files from your local machine available to external services (upload_to_cdn only accepts server-side paths). The returned URL can be passed to remote_deploy as archive_url.

// 输出示例(content[0].text)
{"cdn_url":"https://mock.example.com/2r6hwi8st5gg/file","status":"ok"}

Upload a file to CDN and return the public CDN URL. **IMPORTANT: When sending files via messaging platforms (Telegram, Slack, etc.), you MUST use this tool first to get a CDN URL.** Local file paths are NOT accessible outside the agent environment. Only CDN URLs can be accessed by external services. Returns: cdn_url: Public CDN URL

// 输出示例(content[0].text)
{"cdn_url":"https://mock-cdn.example.com//tmp/i.png","status":"ok"}

Deploy a directory of individual HTML slide files as a unified presentation with navigation controls. This tool validates image references, combines slide HTML files into a single presentation viewer HTML, and deploys it. Requirements: - slides_dir must be an absolute path containing slide HTML files and an optional imgs/ subdirectory - All images must be under an 'imgs/' subdirectory - Image references must use bare relative paths (e.g. 'imgs/photo.png'), NOT absolute or '../' paths - HTTP/HTTPS URLs and data URIs are allowed Slide ordering: - If html_files is provided, slides are ordered as listed - If omitted, all .html files are sorted by natural order (slide2.html < slide10.html)

// 输出示例(content[0].text)
{"slides_dir":"/tmp/slides","status":"ok","url":"https://mock.example.com/slides"}

Initialize a React project (Vite + React + TypeScript + TailwindCSS). All dependencies are installed automatically.

// 输出示例(content[0].text)
{"project_name":"p","status":"ok","target_dir":"/tmp/p"}