Venice MCP Server | Venice API Docs

The Venice MCP Server is the official Model Context Protocol server for Venice. It exposes the Venice API (chat, image, video, audio, music, embeddings, web augment, characters, crypto RPC, and TEE attestation) plus account operations (billing and API keys) as 45 tools that any MCP-compatible agent can call.

GitHub: veniceai/venice-mcp-server

Published as @veniceai/mcp-server on npm. MIT licensed.

45 Tools

Every Venice modality plus billing, API-key, and TEE tooling

Any MCP Host

Claude Desktop, Cursor, ChatGPT, LM Studio, Continue, and more

Wallet Auth (Optional)

Bring an API key, or pay per call with a SIWE-signed wallet via x402

Quick start

Get a Venice API key

Generate one from venice.ai/settings/api. See the API key guide for step-by-step instructions.

Add Venice to your MCP host config

Drop this into your MCP host’s config file:

{
  "mcpServers": {
    "venice": {
      "command": "npx",
      "args": ["-y", "@veniceai/[email protected]"],
      "env": { "VENICE_API_KEY": "<your-venice-api-key>" }
    }
  }
}

Common config paths:

Host	Path
Claude Desktop (macOS)	`~/Library/Application Support/Claude/claude_desktop_config.json`
Claude Desktop (Windows)	`%APPDATA%\Claude\claude_desktop_config.json`
Cursor	`~/.cursor/mcp.json`
LM Studio	`mcp.json` (from the app’s MCP settings)

Restart your MCP host

Your agent now has chat, image, video, music, TTS, ASR, and 39 more Venice tools available.

Most MCP hosts only pass environment variables that are explicitly listed in the env block. System-level env vars are not inherited. If you see 402 errors with an API key set, double-check that VENICE_API_KEY is inside env in your config.

What you get

45 tools spanning every Venice modality plus account operations, 3 resources (venice://models, venice://styles, venice://voices), and 3 prompt templates (uncensored research, NSFW creative writing, image style explorer). Every tool declares explicit readOnlyHint, destructiveHint, idempotentHint, and openWorldHint annotations, so hosts can run lookups without prompting and ask for confirmation before anything destructive, such as media cleanup or a crypto relay.

Chat & embeddings

Tool	Description
`venice_chat`	OpenAI-compatible chat completion against Venice’s full LLM catalog, with text/image/audio/video/file blocks, structured outputs, function tools, prompt caching, and reasoning controls. Calls are plaintext and non-streaming.
`venice_responses`	Beta, stateless OpenAI-compatible Responses API with text/image input and reasoning controls. This tool doesn’t expose tool-calling fields, and E2EE models aren’t supported.
`venice_embeddings`	Compute embeddings for text input.
`venice_chat_with_character`	Chat with a Venice character by slug.

Image

Tool	Description
`venice_image_generate`	Generate an image (Flux 2, Lustify SDXL, Anime/WAI, Qwen Image, GPT Image, Nano Banana Pro, and more), with aspect ratio or width/height, quality tiers, prompt enhancement, style references, web search, and variants.
`venice_image_edit`	Edit an image with a prompt.
`venice_image_multi_edit`	Edit up to 8 images together with one prompt (composition / outpainting).
`venice_image_upscale`	Upscale an image 2–4×, with a `creativity` control.
`venice_image_remove_bg`	Remove an image background.
`venice_image_styles`	List image style presets.

Video

Tool	Description
`venice_video_generate`	Queue a video generation (Sora 2, Veo 3.1, Kling, Wan, LTX 2, Seedance, Runway Gen-4, and more) with image, video, audio, reference, and element inputs.
`venice_video_status`	Check status of a queued video job (`POST /video/retrieve`).
`venice_video_complete`	Mark a completed video as downloaded and delete its server-side media. Reports the deletion only when Venice confirms it.
`venice_video_quote`	Get a price quote before queuing.

Audio (TTS / ASR)

Tool	Description
`venice_tts`	Text-to-speech with cloned voices, temperature, and emotion tags. Streaming is buffered into one complete audio result.
`venice_asr`	Transcribe audio from a URL, with optional word/character timestamps. Large transcripts are paged through a session handle.
`venice_voice_clone`	List live model-scoped voices, or clone a voice from a sample URL (returns a `vv_<id>` handle for `venice_tts`).
`venice_audio_quote`	Get a price quote for music generation.

Music

Tool	Description
`venice_music_generate`	Queue music generation (`ace-step-15`, `elevenlabs-music`, `minimax-music-v2/v25/v26`, `stable-audio-25`, `mmaudio-v2-text-to-audio`, `elevenlabs-sound-effects-v2`) with lyrics, instrumental, loop, voice, language, speed, and duration controls.
`venice_music_status`	Check status of a queued music job.
`venice_music_complete`	Mark a completed music job as downloaded.

Web augment

Tool	Description
`venice_web_search`	Search the web with Brave (default, Zero Data Retention) or Google (proxied and anonymized by Venice).
`venice_web_scrape`	Scrape one URL into markdown.
`venice_text_parser`	Extract text from PDF/DOCX/EPUB/PPTX/XLSX.

Catalog & characters

Tool	Description
`venice_list_models`	Page through the live model catalog of every type, with compact summaries, prices, and capability flags.
`venice_model_details`	Get one exact model’s full row: `model_spec` constraints, capabilities, and pricing.
`venice_model_traits`	Map trait names (`default`, `most_uncensored`, …) to current model IDs.
`venice_model_compatibility_mapping`	Map compatible model names to Venice model IDs.
`venice_list_characters`	List public Venice characters with search, tag, model, and sort filters. API key only.
`venice_get_character`	Get one public character by slug. API key only.
`venice_character_reviews`	List paginated public reviews for a character. API key only.

TEE attestation

Tool	Description
`venice_tee_attestation`	Fetch hardware-attestation evidence (Intel TDX quote, optional NVIDIA evidence, signing key) for a TEE-capable text model, using your own 32-byte nonce.
`venice_tee_signature`	Fetch the enclave signature payload for a completed TEE chat request ID.

Both TEE tools return provider evidence only: they do not verify it, and they do not encrypt chat. Verify the echoed nonce, debug-mode state, and signing-key binding yourself before trusting a TEE model; see Verifying TEE attestation. enable_e2ee is rejected by this server.

Crypto

Tool	Description
`venice_crypto_networks`	List the live network slugs accepted by the RPC proxy.
`venice_crypto_rpc`	Proxy one JSON-RPC request, or a batch of up to 100, to Ethereum, Base, Arbitrum, Optimism, Polygon, Solana, Starknet, zkSync, Linea, BSC, Avalanche, Blast, and more.

Transaction broadcasts (eth_sendRawTransaction, eth_sendUserOperation, Solana sendTransaction, Starknet writes) must be sent as single requests with an idempotency_key, and the same key must be reused on retry.

Billing (admin keys)

Tool	Description
`venice_billing_balance`	Current USD, DIEM, and bundled-credit availability.
`venice_billing_usage_analytics`	Beta aggregated usage by date, model, and API key (lookback or custom range).
`venice_billing_usage_history`	Detailed usage with cursor pagination and JSON or CSV output.

The billing tools require an ADMIN API key. Inference keys cannot call these endpoints, and the server fails locally instead of falling back to a wallet token.

API keys

Tool	Description
`venice_list_api_keys`	List active key metadata, including only the last six characters. Admin key only.
`venice_get_api_key`	Get one key’s metadata, usage, balances, and rate limits by key ID. Admin key only.
`venice_api_key_rate_limits`	Current balances, access status, tier, expiration, and model-specific rate limits. Inference or admin key.
`venice_api_key_rate_limit_logs`	The last 50 exceeded rate-limit events (experimental). Admin key only.

x402 wallet helpers

Only relevant if you authenticate with a wallet via x402 instead of an API key.

Tool	Description
`venice_x402_balance`	Check prepaid x402 credit balance for an EVM or Solana wallet address.
`venice_x402_top_up_info`	Fetch top-up requirements (network, USDC token, receiver, min amount). Takes no arguments.
`venice_x402_transactions`	List recent x402 top-up and debit transactions for an EVM or Solana wallet address.

Configuration

The server is configured entirely through environment variables.

Models and requests

Env var	Default	Notes
`VENICE_API_KEY`	(none)	Your Venice API key. The simplest setup.
`VENICE_DEFAULT_CHAT_MODEL`	`deepseek-v4-flash-0731`
`VENICE_DEFAULT_IMAGE_MODEL`	`flux-2-pro`
`VENICE_DEFAULT_TTS_MODEL`	`tts-kokoro`
`VENICE_DEFAULT_ASR_MODEL`	`openai/whisper-large-v3`
`VENICE_DISABLE_NSFW`	`0`	Set to `1` to remove NSFW capability notes from tool descriptions.
`VENICE_HTTP_TIMEOUT_MS`	`60000`
`VENICE_SIWX_TOKEN`	(none)	x402 wallet-mode auth token. See x402 below.

If both VENICE_API_KEY and VENICE_SIWX_TOKEN are set, the API key wins on most calls; see Auth-mode coverage for the exceptions.

Response size limits

Generated media is buffered and embedded in the MCP result as base64, so each modality has a byte ceiling.

Env var	Default	Applies to
`VENICE_MAX_IMAGE_RESPONSE_BYTES`	`33554432` (32 MiB)	`venice_image_generate`, `venice_image_edit`, `venice_image_multi_edit`, `venice_image_upscale`, `venice_image_remove_bg`
`VENICE_MAX_VIDEO_RESPONSE_BYTES`	`26214400` (25 MiB)	`venice_video_status`
`VENICE_MAX_AUDIO_RESPONSE_BYTES`	`33554432` (32 MiB)	`venice_tts`, `venice_music_status`

An oversized image or TTS result is discarded with an error. An oversized video or music result stays queued, so you can call the status tool again with the same queue ID after raising the limit.

HTTP server

Only relevant in the Streamable HTTP deployment.

Env var	Default	Notes
`PORT`	`3333`	HTTP listener port.
`VENICE_MCP_HOST`	`127.0.0.1`	Bind address. Set to `0.0.0.0` for LAN or container exposure.
`VENICE_MCP_AUTH_TOKEN`	(none)	Bearer token for `/mcp`. Required when bound outside loopback, where it must be at least 16 characters. Once set, it’s enforced on loopback too.
`VENICE_MCP_ALLOW_UNAUTHENTICATED_HTTP`	`0`	Escape hatch for exposed unauthenticated HTTP. Use only behind a trusted authenticated proxy.
`VENICE_MCP_MAX_SESSIONS`	`100`	Maximum active Streamable HTTP sessions.
`VENICE_MCP_SESSION_TTL_MS`	`1800000`	Idle session lifetime before cleanup.

Media generation behavior

Completed media can be bytes or a URL. Video and music jobs return embedded MP4/audio resources, or JSON with a download_url that comes back as a resource link. Download that link first, then call venice_video_complete or venice_music_complete; media cleanup is never applied to a download_url result because the link stops working once the file is deleted.
Some models only return a URL at queue time. For VPS and Grok Imagine Private models, capture the download_url returned by venice_video_generate and pass it into venice_video_status; a COMPLETED retrieve without an inline URL still yields a usable link.
Prompt enhancement is reported back. When enhance_prompt is applied, image generate, edit, and multi-edit results include the URL-decoded prompt Venice returned.
ASR paging is charged once. Transcripts longer than 8,000 characters, or with more timestamps than fit on one page (200 per array by default, up to 500 with timestamp_limit), return a result_handle with the first page. Request later pages with that handle. Handles live in memory per MCP session for 10 minutes, and only the 16 most recent are kept.
Seedance consent is not a policy bypass. Current public Seedance models may reject media containing detectable persons outright. Where a needs_consent response is returned, the three consents.seedance flags are legal attestations: set them to true only after the user explicitly confirms each statement.

x402 wallet mode

Venice supports authenticating with a Sign-In-With-X wallet token backed by prepaid USDC credit on Base or Solana, in addition to the normal API key flow. No email, phone, or KYC required: your wallet is the only identity.

{
  "mcpServers": {
    "venice": {
      "command": "npx",
      "args": ["-y", "@veniceai/[email protected]"],
      "env": { "VENICE_SIWX_TOKEN": "<base64 Sign-In-With-X payload>" }
    }
  }
}

The MCP server forwards VENICE_SIWX_TOKEN in the SIGN-IN-WITH-X header on every Venice API call. The server never sees your private key. Wallet signing and USDC top-up authorizations happen in your own wallet.

Flow	What happens
One-time setup	Sign a Sign-In-With-X message in your wallet → produces a SIWX token (base64 JSON).
Top up	`POST /api/v1/x402/top-up` returns 402 + payment requirements. Sign a USDC payment for one of the returned Base or Solana options, resubmit, and Venice credits your balance.
Every inference call	MCP server sends `SIGN-IN-WITH-X: <SIWX>`; Venice debits your prepaid balance.

Minimum top-up is $5 USD. Minimum balance to call inference is $0.10. Once topped up, calls are sub-100ms because settlement happens off-chain on a fast credit account.

Wallets linked to a Venice account with DIEM staked consume from the staking balance instead of USDC credits, so no top-up is needed.

Venice does not accept payment headers on inference routes, so this server never sends one. PAYMENT-SIGNATURE belongs only to /x402/top-up, and the top-up helper discovers requirements without submitting a payment.

Auth-mode coverage

Some endpoints accept only one auth mode:

Tool	API key	x402
Inference, media, web augment, crypto RPC, character chat	✓	✓
`venice_list_characters`, `venice_get_character`, `venice_character_reviews`	✓	✗
`venice_billing_*`, `venice_list_api_keys`, `venice_get_api_key`, `venice_api_key_rate_limit_logs`	✓ (admin)	✗
`venice_api_key_rate_limits`	✓ (any key)	✗
`venice_x402_balance`, `venice_x402_transactions`	✗	✓
`venice_x402_top_up_info`	✓	✓

When both credentials are set, the API key is used for every authenticated call except venice_x402_balance and venice_x402_transactions, which always send the wallet token.

Self-hosting (Streamable HTTP)

For team or workspace deployments, run the MCP server over HTTP instead of stdio:

docker run -p 3333:3333 \
  -e VENICE_API_KEY=<your-venice-api-key> \
  -e VENICE_MCP_AUTH_TOKEN=<choose-a-long-random-token> \
  ghcr.io/veniceai/venice-mcp-server:latest

The server is now available at http://localhost:3333/mcp. HTTP clients must send Authorization: Bearer <VENICE_MCP_AUTH_TOKEN> and should create new sessions without an mcp-session-id header, then reuse the server-issued session ID.

/mcp is a credential-backed tool execution endpoint: callers can spend the configured Venice API key or x402 balance. When HTTP mode binds outside loopback, startup fails unless VENICE_MCP_AUTH_TOKEN is set, or VENICE_MCP_ALLOW_UNAUTHENTICATED_HTTP=1 is set behind a trusted authenticated proxy. For production, pin the npm package version explicitly instead of relying on latest.

Resources

GitHub

Source code, issues, and releases

npm

@veniceai/mcp-server

Venice Skills

Companion skills that teach agents how to use these tools

MCP Spec

Learn more about the Model Context Protocol

GitHub: veniceai/venice-mcp-server

45 Tools

Any MCP Host

Wallet Auth (Optional)

​Quick start

​What you get

​Chat & embeddings

​Image

​Video

​Audio (TTS / ASR)

​Music

​Web augment

​Catalog & characters

​TEE attestation

​Crypto

​Billing (admin keys)

​API keys

​x402 wallet helpers

​Configuration

​Models and requests

​Response size limits

​HTTP server

​Media generation behavior

​x402 wallet mode

​Auth-mode coverage

​Self-hosting (Streamable HTTP)

​Resources

GitHub

npm

Venice Skills

MCP Spec

Quick start

What you get

Chat & embeddings

Image

Video

Audio (TTS / ASR)

Music

Web augment

Catalog & characters

TEE attestation

Crypto

Billing (admin keys)

API keys

x402 wallet helpers

Configuration

Models and requests

Response size limits

HTTP server

Media generation behavior

x402 wallet mode

Auth-mode coverage

Self-hosting (Streamable HTTP)

Resources