Odel
avots

avots

@avotsaiCommunicationMITUpdated 2w ago

Image, video, audio, face-swap, talking avatars and chat across 300+ AI models, one balance.

Server endpointStreamable HTTP

This is the third-party server itself — Odel doesn't run it. Hitting this URL directly talks straight to the upstream server with no auth or proxying. Connect through Odel to front it with managed auth.

avots-mcp

Official MCP (Model Context Protocol) server for avots.ai - a multi-provider AI platform.

One connection gives you:

  • 🖼 Image generation - Nano Banana (Gemini 2.5 / 3 Pro / 3.1 Flash), GPT-5 Image, FLUX, Recraft, Ideogram
  • 🎬 Video generation - Veo 3.1, Seedance 2.0, Kling v3.0 Pro, Sora 2 Pro, Grok Imagine (async, 1-8 min)
  • 🔄 Face swap & 🗣 talking avatars - swap a face into any video, or make a portrait speak your text (lip-synced)
  • 🎵 Music & voice - ElevenLabs Music, ACE-Step, Stable Audio + text-to-speech (ElevenLabs TTS, preset or cloned voices)
  • 📅 Calendar - turn "meeting tomorrow at 12" into an event on your linked Apple / Google calendar
  • 💬 300+ chat models - Claude (Sonnet / Opus), GPT-5, Gemini 3, DeepSeek, Sonar, and more - billed through one balance

The server lives at https://mcp.avots.ai/ and speaks the MCP 2025-06-18 spec over Streamable HTTP. Tools are billed per call against your avots balance (same balance you'd see on the web app or the Telegram bot).

Quick start

  1. Sign up at avots.ai and mint an MCP key at Settings → Integrations (it looks like av_mcp_<48hex>).
  2. Pick your client from the table below and follow the linked guide.
  3. Try it - ask your client "generate an image of a fox in a snowy forest" and watch tokens get spent.
ClientGuideAuth model
Claude.ai webdocs/claude-web.mdOAuth - paste URL, click Connect, sign in (no token copy-paste)
Claude Desktopdocs/claude-desktop.mdBearer token via mcp-remote
Claude Code (CLI)docs/claude-code.mdBearer token via claude mcp add
Cursordocs/cursor.mdBearer token via mcp-remote
Clinedocs/cline.mdBearer token via mcp-remote
openclawdocs/openclaw.mdNative remote - Bearer header, no bridge
LibreChatdocs/librechat.mdNative remote - Bearer header, no bridge
Continue.devdocs/continue.mdNative remote - Bearer header, no bridge
Any other MCP clientdocs/tools.md - endpoint + tool listBearer header Authorization: Bearer av_mcp_…

Ready-to-paste mcp.json snippets live under examples/.

What's in the server

Ten tools, all listed in docs/tools.md:

ToolCostWhat it does
check_balancefreeCurrent tokens and subscription tier.
list_modelsfreeAll active models with per-call cost (filter by chat, image, video, audio).
chat~10-1000 ⚡Send a prompt to any chat model. Useful for delegating to GPT, DeepSeek, Sonar, etc.
generate_image~200-500 ⚡Synchronous image gen. Returns an inline image block (base64) and a hosted URL.
generate_video~200-5000 ⚡Async video gen with two-step confirmation: first call previews the cost and alternative models; second call (confirmed: true) actually submits.
face_swap_video~500-2000 ⚡Async. Swap a face from a photo into a target video, keeping the original motion (pixverse). Two-step confirm.
generate_talking_avatarvaries ⚡Async. Generate a portrait, speak your text (TTS), and lip-sync it into a talking-head video. Two-step confirm.
generate_audio~50-800 ⚡Music (ElevenLabs Music, ACE-Step, Stable Audio) or spoken voice / TTS (ElevenLabs, preset or cloned voices). Async.
create_calendar_event~5 ⚡Turn natural language into an event on a linked Apple/Google calendar, or return an .ics.
check_jobfreePoll an async job (video / face-swap / avatar) by job_id.

About the two-step video flow. Video is the most expensive tool. To avoid surprise spend, generate_video returns a preview card the first time it's called (no submit, no reserve). The client (e.g. Claude) shows the cost + alternative models with prices and asks the user. The user confirms, the client re-calls with the chosen model + confirmed: true, and only then does the job get submitted. On submit error the server returns the same alternatives card - it never silently swaps to a pricier model.

What you can build

A few things this is actually useful for. Each one chains two or more tools through one connection and one balance.

  • Social ad creative in one prompt — hero image, animated variant, music bed.
  • Product-photo angle pack — one product shot in, four angles + a rotation clip out.
  • Storyboard to animatic — four script-driven frames animated into a 12-second rough cut.
  • Vertical Reels / Shorts factory — 9:16 clip + matching 15-second music bed, repeatable per video.
  • Podcast cover art + show notes — four cover variations and a written episode description for each.
  • Second-opinion delegation — forward a tricky problem to a model from a different lineage and compare.
  • Localized brand assets — translated copy and locale-tuned visuals across markets.

Each of these flows is written out as a runnable script — exact prompt, tool sequence, model picks, cost — in docs/recipes.md.

Cost ranges from ~200⚡ for a single image to ~5000⚡ for a 10-second 1080p Kling Pro clip — run list_models (free) at any time for live per-call prices.

Billing

All tool calls bill against your avots balance, just like the web app and Telegram bot. No separate metering. Daily USD cap (set in Settings) and per-key rate limits apply.

See pricing.avots.ai for token packs and subscriptions.

Troubleshooting

Common cross-client issues (401s, daily caps, two-step video flow, image rendering, npx PATH gotchas) are collected in docs/troubleshooting.md. For client-specific setup, see the per-client guide linked in the table above.

Issues & feedback

Open an issue here, or write to hello@avots.ai. For platform questions (billing, models, web app) the Telegram bot @AvotsAIbot is the fastest channel.

License

MIT - feel free to fork the docs, the examples, and anything else here.