MiniMax MCP Server
November 20, 2026
Table of contents
- useapi.net vs MiniMax’s own MCP server
- Endpoint and authorization
- Setup
- Use it in the Claude app or ChatGPT
- Tools
- Long-running work
- Example prompts
- Limits and notes
The useapi.net MiniMax MCP server lets an AI assistant such as the Claude app, ChatGPT, Claude Code, Cursor or VS Code Copilot make videos, images, songs and speech with Hailuo AI and MiniMax Audio on your own MiniMax accounts. It is a hosted Model Context Protocol server with 38 tools: generate Hailuo, Veo 3.1 and Seedance 2.0 videos from text, frames or omni references, or from a Hailuo AI video template, generate MiniMax H3 Image, Nano Banana, GPT Image, Seedream and Midjourney images, hand a task to the MiniMax agent, make songs and covers, turn text into speech in built-in or cloned voices, upload your own files and clean up your history. Each tool calls the MiniMax API, so the same rules and limits apply. Nothing needs to be installed or hosted.
Is there a MiniMax MCP server? Yes. useapi.net, a third-party service not affiliated with MiniMax, hosts one at https://api.useapi.net/v1/minimax/mcp with 38 tools for Hailuo video, images, music and speech on your own Hailuo AI and MiniMax Audio subscriptions, nothing to install. MiniMax’s own MCP server has 8 tools, needs a developer API key at pay-as-you-go prices and has no tools for songs, covers, Veo, Seedance, templates or the agent (comparison).
Prefer REST? Code: github.com/useapi/minimax-api
See all useapi.net MCP servers on Model Matrix & MCP.
useapi.net vs MiniMax’s own MCP server
MiniMax also publishes its own MiniMax-MCP server with 8 tools for speech, voices, Hailuo video and MiniMax images. You install and run it yourself with a MiniMax developer API key, billed at developer platform prices. The useapi.net server is hosted and works on your Hailuo AI and MiniMax Audio subscriptions (as of November 20, 2026):
| useapi.net MiniMax MCP | MiniMax’s own MCP | |
|---|---|---|
| Nothing to install, connects in the Claude app and ChatGPT | Yes | No, you run it with uvx or host it yourself |
| Billed on your Hailuo AI and MiniMax Audio subscriptions | Yes | No, a developer API key at pay-as-you-go prices |
| Songs, and covers of a reference track | Yes | No |
| Veo 3.1 and Seedance 2.0 videos besides Hailuo | Yes | No, Hailuo only |
| Nano Banana, GPT Image, Seedream and Midjourney images besides MiniMax H3 | Yes | No, MiniMax H3 Image and image-01 only |
| Hailuo AI video templates and the MiniMax agent | Yes | No |
| List and delete your videos, images, songs, speech and uploads | Yes | No |
| Several MiniMax accounts in one connection | Yes | One API key per server |
useapi.net is a third-party service, not affiliated with MiniMax.
Endpoint and authorization
https://api.useapi.net/v1/minimax/mcp
- Transport: Streamable HTTP.
- Listed in the official MCP Registry as
net.useapi/minimax, so MCP clients and directories that read the registry can find it by that name. - Authorization: your useapi.net API token in the
Authorization: Bearer …header, see Setup useapi.net. The Claude app and ChatGPT sign in with OAuth instead, see Use it in the Claude app or ChatGPT. - You need at least one MiniMax account added with Setup MiniMax. Adding and removing accounts is not available over MCP: use POST accounts or the setup page.
Setup
Replace YOUR_API_TOKEN with your useapi.net API token.
Claude Code
claude mcp add --transport http useapi-minimax https://api.useapi.net/v1/minimax/mcp --header "Authorization: Bearer YOUR_API_TOKEN"
Cursor
In ~/.cursor/mcp.json:
{
"mcpServers": {
"useapi-minimax": {
"url": "https://api.useapi.net/v1/minimax/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_TOKEN"
}
}
}
}
VS Code
In .vscode/mcp.json:
{
"servers": {
"useapi-minimax": {
"type": "http",
"url": "https://api.useapi.net/v1/minimax/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_TOKEN"
}
}
}
}
Windsurf and other clients
Any client that supports remote MCP servers over HTTP with custom headers works. The usual JSON shape:
{
"mcpServers": {
"useapi-minimax": {
"serverUrl": "https://api.useapi.net/v1/minimax/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_TOKEN"
}
}
}
}
Some clients name the address field url instead of serverUrl.
Test it with curl
curl -s https://api.useapi.net/v1/minimax/mcp \
-H "Authorization: Bearer YOUR_API_TOKEN" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"curl","version":"1"}}}'
curl -s https://api.useapi.net/v1/minimax/mcp \
-H "Authorization: Bearer YOUR_API_TOKEN" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'
The first call answers with the server’s name useapi-minimax and its instructions for the assistant, the second with the 38 tools.
Use it in the Claude app or ChatGPT
The Claude app (claude.ai and Claude Desktop) and ChatGPT can add this server as a custom connector. They sign in with OAuth: you give the app the URL, it opens a useapi.net page, and you paste your useapi.net API token there once. Both work on their Free plans. The API-token setups above for Claude Code, Cursor, VS Code and other clients keep working as before.
https://api.useapi.net/v1/minimax/mcp
Claude app
- Open Settings → Connectors and click Add custom connector.
- Enter a name, for example
useapi MiniMax, and the URL above, then click Add. - Click Connect. Claude opens the useapi.net sign-in page.
- Paste your useapi.net API token and click Allow.
Claude asks before each tool call unless you choose Always allow. The Free plan allows one custom connector.
ChatGPT
- Open Plugins, click Add and choose Add custom MCP server.
- Enter a name and the URL above, and set Authentication to OAuth.
- Tick ChatGPT’s “I understand and want to continue” box, then click Create as a plugin.
- Click Connect. ChatGPT opens the useapi.net sign-in page.
- Paste your useapi.net API token and click Allow.
ChatGPT asks you to confirm each tool call that creates or changes something. The Free plan works without developer mode (as of October 25, 2026).
How the sign-in works
- Your API token is in your welcome email, see Setup useapi.net. The useapi.net page links there too.
- Each connection is for one MCP server. Connect each server you want to use on its own.
- The app gets an access token that lasts 1 hour and renews it on its own. A connection you don’t use for 90 days lapses, and you connect again.
- Resetting your API token disconnects every app connected with it.
- Results come back as public links, so they open right in the app.
Tools
Accounts are named by their account id, as minimax_list_accounts shows it. Every id a tool returns (videoId, imageId, fileID, musicId, audioId, voice_id) names its account, so a tool that works on one runs on that account. Without account, a video goes to one of your accounts with a free slot and a song to the account with the fewest songs running. Images, speech, the agent, uploads and the tools that read one account’s lists need account only when more than one account is added. Every create tool also takes wait, and a webhook with replyUrl and replyRef.
Accounts and credits
| Tool | What it does | Main parameters |
|---|---|---|
minimax_list_accounts | Lists your accounts as stored: account id, host, maxJobs, the features each one supports (video, audio, music, agent), session expiry and any stored account error, without calling MiniMax. | |
minimax_get_account | One account’s stored settings: host, maxJobs, supported features, session times and the stored error if MiniMax put it on hold, without calling MiniMax. | account |
minimax_get_features | Reads an account live from MiniMax: its video and image credits, plan and per-model settings, plus the separate MiniMax Audio credits for music and speech. | account |
The parameters match GET accounts, GET accounts/account and GET features.
Uploads, files and characters
| Tool | What it does | Main parameters |
|---|---|---|
minimax_upload_file | Uploads your own image (PNG, JPEG or WebP), audio (MP3 or WAV) or MP4 video, up to 50 MB, from a public https URL or sent as contentBase64, as a reference for videos, images and template videos. Returns a fileID that works on that account only, and its ossPath. | url or contentBase64, mimeType, account |
minimax_list_files | Lists the files uploaded to one account, here, through the REST API or in the MiniMax web app, with each file_id, name, type, CDN URL and oss_path. | account, limit |
minimax_delete_file | Deletes one uploaded file. Generations that already used it are not affected. It cannot be undone. | fileID |
minimax_list_characters | Lists the characters saved on one Hailuo AI account, with each character image’s fileID for minimax_create_video (such as the S2V-01 subject) or minimax_create_template_video, page by page. | account, page, pageSize |
The parameters match POST files, GET files, DEL files/fileID and GET videos/characters. Files larger than 50 MB, up to 5 GB, go to POST files directly.
Videos
| Tool | What it does | Main parameters |
|---|---|---|
minimax_create_video | Generates a video with Hailuo 3.0, 2.3, 02 or 01, Veo 3.1 or Seedance 2.0: from text, a first frame, an end frame, subject images on the S2V models, or omni references cited in the prompt as @Image1, @Video1 or @Audio1 on Seedance and Hailuo 3.0. Hailuo 02, Hailuo 2.3 and Veo set size and length with options, Seedance and Hailuo 3.0 with resolution and duration. | prompt, model, options, resolution, duration, aspectRatio, fileID, end_frame_fileID, fileID2 to fileID14, videoFileID1 to videoFileID3, audioFileID1 to audioFileID3, account, maxJobs |
minimax_get_video | Reads one video, and with wait: true waits for it to finish. A finished video comes with its videoURL, a downloadURL without watermark on paid accounts, its cover, width and height. A 404 usually means the video was moderated and removed. | videoId, wait |
minimax_list_videos | Lists one account’s videos on Hailuo AI, including running ones and those made in the web app, page by page. | account, limit, lastVideoId |
minimax_cancel_video | Asks MiniMax to cancel a video that is still generating. minimax_get_video then shows its final status. It cannot be undone. | videoId |
minimax_delete_video | Deletes one video from its account’s Hailuo AI history. It cannot be undone. | videoId |
The parameters match POST videos/create, GET videos/videoId, GET videos, POST videos/cancel and DEL videos/videoId, with the same names for the numbered references. Durations, resolutions and ratios of each model are on POST videos/create.
Images
| Tool | What it does | Main parameters |
|---|---|---|
minimax_create_image | Generates 1 to 4 images with MiniMax H3 Image, image-01, Nano Banana, GPT Image, Seedream or Midjourney (always 4), from a prompt and up to 16 reference images cited as @Image1 on the models that support it, or a character image on image-01. The first image comes back as an image the assistant can see. | prompt, model, aspectRatio, resolution, quality, quantity, referenceFileId1 to referenceFileId16, characterFileId, account |
minimax_get_image | Reads one image, and with wait: true waits for it to finish. A finished image comes with its URL, a downloadURL without watermark on paid accounts, its size and a fileID to reuse as a reference, and as an image block. | imageId, wait |
minimax_list_images | Lists one account’s images on Hailuo AI, including running ones and those made in the web app, page by page. Deleted images are not listed. | account, limit, lastImageId |
minimax_delete_image | Deletes one image from its account’s Hailuo AI history. It leaves the list, but minimax_get_image can still read it. It cannot be undone. | imageId |
The parameters match POST images/create, GET images/imageId, GET images and DEL images/imageId.
Template videos and the agent
| Tool | What it does | Main parameters |
|---|---|---|
minimax_list_agent_templates | Lists the Hailuo AI video templates with each one’s id, title and preview, page by page. With templateId, one template’s credits and the prompt or image it requires. | templateId, limit, lastId, account |
minimax_create_template_video | Generates a video from a template, with the prompt or image the template requires. It returns a videoId for minimax_get_video. | templateId, prompt, fileID, ossPath, account, maxJobs |
minimax_run_agent | Gives the Hailuo AI agent a task in plain words and the models it may use, such as Hailuo 2.3, Veo 3.1, Nano Banana 2 or Midjourney V7, with up to 3 reference files by public URL. The agent picks among them and returns the files it made. | prompt, models, fileUrls, account |
minimax_get_agent_job | Reads one agent job, and with wait: true waits for it to finish. A finished job comes with each file the agent made. Jobs are kept 31 days. | jobId, wait |
minimax_list_agent_jobs | Lists the agent jobs running now, by account, and the accounts that can run the agent. |
The parameters match GET videos/agent-templates, POST videos/agent-create, POST agent, GET agent/jobId and GET agent/jobs.
Music
| Tool | What it does | Main parameters |
|---|---|---|
minimax_create_music | Generates 1 to 3 songs with MiniMax Music 3.0 or 2.6 from a style prompt, your own lyrics or a reference track (a cover). Without lyrics the model writes them, and instrumental: true makes a song without vocals. | prompt, lyrics, title, model, quantity, instrumental, referenceMusicId, account |
minimax_upload_music_reference | Uploads an MP3 or WAV of up to 50 MB as a reference track for covers. Returns its musicId, ready at once, and the lyrics MiniMax transcribed from it. A reference stays on the account for any number of covers, and uploading it costs no credits. | url or contentBase64, mimeType, name, account |
minimax_get_music | Reads one song or reference, and with wait: true waits for a song to finish. A finished song comes with its MP3 audio_url, duration, title, lyrics and model. | musicId, wait |
minimax_list_music | Lists one account’s songs, references or both, including songs made on minimax.io and tracks still rendering, page by page. | account, type, limit, lastMusicId, favorite |
minimax_delete_music | Deletes one song or reference track. Its audio_url stops working, so download it first. A song still rendering cannot be deleted. It cannot be undone. | musicId |
The parameters match POST music/create, POST music/reference, GET music/musicId, GET music and DELETE music/musicId.
Speech and voices
| Tool | What it does | Main parameters |
|---|---|---|
minimax_create_speech | Turns text into speech with MiniMax Speech 2.8 HD (the default) or another Speech 2.8, 2.6, 2.5, 02 or 01 model, HD or Turbo, up to 5,000 characters on HD models and 10,000 on Turbo, in a built-in or cloned voice, with speed, volume, pitch, emotion and voice effects. | text, voice_id, model, speed, vol, pitch, emotion, language_boost, account |
minimax_get_speech | Reads one recording, and with wait: true waits for it to finish. A finished one comes with its MP3 audio_url, text, model, voice, settings and the credits MiniMax charged. An audioId from minimax_create_speech resolves for 7 days. | audio_id, wait |
minimax_list_speech | Lists one account’s recordings, made here or on minimax.io, page by page. | account, page, page_size |
minimax_delete_speech | Deletes one recording. One still generating cannot be deleted. It cannot be undone. | audio_id |
minimax_list_voices | Lists MiniMax’s built-in voices, filtered by tags and page by page, or your cloned voices, each with its voice_id and a sample. | account, is_system, tag_list, page, page_size |
minimax_get_speech_config | Lists the values speech accepts on an account: the emotions, and the voice tags by language, gender, age and accent. | account |
minimax_get_voice_quota | Shows how many cloned voice slots an account has, how many are in use and the free clone previews left. | account |
minimax_upload_voice_sample | Uploads one MP3 or WAV sample of 10 to 60 seconds of speech, up to 50 MB, for voice cloning. Returns a fileID for minimax_clone_voice. | url or contentBase64, mimeType, account |
minimax_clone_voice | Builds a new voice from 1 to 10 samples of the same account and returns its voice_id and a sample. It takes a cloned voice slot. | files, voice_name, language_tag, need_noise_reduction |
minimax_delete_voice | Deletes one cloned voice and frees its slot. It cannot be undone. | voice_id |
The parameters match POST speech/create, GET speech/audio_id, GET speech, DEL speech/audio_id, GET speech/voices, GET speech/config, GET speech/equity, POST speech/upload-sample, POST speech/clone-voice and POST speech/delete-voice. Speech streamed over a WebSocket or as server-sent events stays on the REST API.
Running jobs
| Tool | What it does | Main parameters |
|---|---|---|
minimax_list_running_jobs | Lists the videos, images, music and speech the API is tracking now, and the accounts with a free video slot (maxJobs, running and free). | |
minimax_stop_tracking | Stops tracking a running video, image, music or speech job and frees its slot on the account. It does not cancel the work at MiniMax: use minimax_cancel_video for that. It cannot be undone. | id |
The parameters match GET scheduler, GET scheduler/available and DEL scheduler/id.
The full input schema of every tool comes from tools/list.
Long-running work
MCP clients usually give up on a tool call after about a minute, so a tool waits up to about 50 seconds:
- A quick result can come back from the create tool itself.
- Otherwise the tool returns the id, and the matching get tool (
minimax_get_video,minimax_get_image,minimax_get_agent_job,minimax_get_musicorminimax_get_speech) withwait: truepicks it up. Ask the assistant to keep checking until it is done. wait: falseon a create tool returns the id at once.
Clients that ask for progress notifications get one about every 10 seconds while a tool waits.
Videos, images, songs and speech come back as MiniMax CDN links.
Example prompts
- “Make a 6-second Hailuo 2.3 video at 768p of a paper boat drifting down a rainy street at night.”
- “Upload this photo and make a 10-second Hailuo 3.0 video in 2K where @Image1 walks along a beach at sunset.”
- “Generate Midjourney V7 images of a cozy mountain cabin in winter.”
- “Show me a few Hailuo video templates, then make one with my photo.”
- “Write an upbeat synth-pop song about a summer road trip with Music 3.0.”
- “Upload this MP3 and make a cover of it that keeps its lyrics.”
- “Clone my voice from these two recordings, then read this paragraph with it.”
- “How many video and audio credits are left on my MiniMax account?”
Limits and notes
- The tools use your MiniMax accounts and the same rules as the REST API, such as
maxJobsand one song at a time per account, songs started on minimax.io included. Videos and images use the account’s Hailuo AI credits, songs and speech the separate MiniMax Audio credits, see Pricing, andminimax_get_featuresshows both. - A 412 means the account lacks the credits for that job. A 429 means your accounts are busy or MiniMax’s own queue is full: check
minimax_list_running_jobsand retry later. A 596 means MiniMax put the account on hold. - A client may send up to 20 JSON-RPC requests in one batch.
- If the client sends an
MCP-Protocol-Versionheader, it must be one of the supported versions:2025-11-25,2025-06-18,2025-03-26or2024-11-05.