Page through the live Carpe Diem model catalogue (Venice models plus Carpe metadata). Filter by type (text, image, video, tts, …), capability, privacy or a search string. Use carpe_diem_type to pick the right tool: text/code → carpe_chat, embedding → carpe_embeddings, image → carpe_image_generate, imageEdit → carpe_image_edit, upscale → carpe_image_upscale, tts → carpe_tts, asr → carpe_transcribe, music → carpe_music_generate, video/imageToVideo/referenceToVideo → carpe_video_generate. route "router" reads /router/models, which adds the best-price markets able to serve each text model. Free.

Everything about one model: catalogue row (capabilities, constraints such as video durations or music lengths, TTS voices), Carpe price (USD per 1M tokens, or fixed cost per image/call), Venice rate limits and live capacity. Free.

Live Carpe prices. Token models: USD per 1M input/output tokens (inputPrice/outputPrice, after the dynamic multiplier; inputBase/outputBase are Venice list prices). Media: fixed USD per image/edit/web call. Filter by model substring. Free.

Price a text request BEFORE running it, on both surfaces: what /v1 (Carpe alone, in its TEE) would bill, and what /router (best price) would pick among the markets — with each market's price, availability and whether it is a TEE. Give messages (or a prompt) so the input size is estimated, or input_tokens directly. Costs nothing.

Live state of the /router best-price markets: which markets are on, the external routing fee, observed latency per market, and whether external execution is open to this key. Free.

Operator health (database, settlement, providers) and marketplace capacity (headroom, dynamic price multiplier). Free.

Spendable balance of the Carpe Diem account behind CARPE_API_KEY: credits (1 credit = $0.01; escrow = purchased, pending = spent not yet settled, holds = reserved for in-flight requests, available = spendable), unsettled debt, prepaid account and the active payment rail. Free.

Per-request history of the account, newest first: model, provider pseudonym, tokens, cost (USDC), market that served, routing reason, latency, savings. Free.

Account spend aggregated by model and by day over the last N days (1–90, default 7). Free.

Requests, tokens and cost per API key of the account over the last N days, with each key's rail and status. Free.

How the account's /router requests were served: wins per market and savings versus Carpe-only pricing. Free.

The operator rate limit applied to this key right now (max per window, remaining, reset), separate read budget, and parallel lanes. Free.

Credit purchases made by card on the account. Free.

Card top-up for this account. Nothing is charged by this tool: with create: true it returns a checkout URL that a HUMAN must open to pay (valid for a limited time); otherwise it only lists the credit tiers. Never open or pay the link yourself.

What this MCP server has spent since it started, per tool, against the CARPE_MAX_SPEND_USD and CARPE_MAX_PER_CALL_USD guards. Local, free.

x402 mode only: the agent wallet address, its USDC balance on Base, the permit budget, what was billed and what is left. Free.

IRREVERSIBLE: permanently revokes the API key this server uses (CARPE_API_KEY). Every later call with it fails. Use only if the user explicitly asks, or if the key leaked. Requires confirm: "REVOKE".

Run a chat completion on Carpe Diem (OpenAI-compatible; Venice models: GPT, Claude, Gemini, DeepSeek, GLM, Qwen, Llama, uncensored models…). Two surfaces: route "carpe" (/v1, Carpe's TEE, private) or route "router" (/router, cheapest market). Returns the answer plus usage, the exact cost billed and which market served it. Supports function tools, response_format (JSON schema), reasoning_effort and Venice web search. Paid — use carpe_quote first for expensive requests.

Compute embedding vectors (OpenAI-compatible) for 1–128 texts. route "carpe" (/v1) or "router" (/router, cheapest market; privacy applies). Paid per input token.

Live web search through Carpe (Venice augment/search). Always on /v1 — no market arbitrates web search. Flat fee per call (see carpe_pricing, model augment-search).

Fetch one web page as text/markdown through Carpe (Venice augment/scrape). Always on /v1. Flat fee per call.

Generate image(s) from a prompt (Venice image models: Flux, GPT Image, Nano Banana, Seedream, Qwen Image, Recraft, uncensored SD…). Images are returned inline and/or saved to CARPE_OUTPUT_DIR. Fixed price per image (carpe_pricing). Slow models run as a queued job automatically.

Transform one image from a prompt (imageEdit models such as qwen-image-2-edit, nano-banana-2-edit, gpt-image-2-edit). image: https URL, data: URI or (stdio) a local file path.

Compose or edit from 1–3 source images and a prompt (e.g. "put the product from image 1 on the table of image 2"). Always on /v1 — no market arbitrates multi-edit. Runs as a queued job.

Increase image resolution 1–4× (source ≥ 256×256). Always on /v1. image: https URL, data: URI or local path (stdio).

Fetch the result of a queued image job (job_id from an image tool). Polls on the surface that created the job. Free.

Turn text (≤ 4096 characters) into speech. Voices are model-specific: omit voice for the model default, or read the voices of a model with carpe_model_details. A cloned voice (vv_… from carpe_voice_clone) only works on route "carpe" with the model it was cloned for. Billed per character.

Speech to text (Whisper-compatible). audio: https:// URL, data: URI, or local path (stdio). Formats such as mp3, wav, m4a, flac, ogg, webm.

Upload a short reference sample (one speaker) and get a vv_… voice handle for carpe_tts (route "carpe", same model). The upload itself is free but needs a funded account. ONLY with consent: the voice must be the user's own, or its speaker must have agreed to be cloned. The handle expires (~7 days).

Queue a video (Sora, Veo, Kling, Seedance, Wan, LTX, Grok Imagine…; text-to-video, image-to-video, reference-to-video). The payload is priced first (free quote) and refused locally above CARPE_MAX_PER_CALL_USD. Returns a job_id; carpe_video_status downloads the MP4 when ready (videos take minutes). Read the model constraints (durations, aspect ratios, resolutions) with carpe_model_details.

Check a video job from carpe_video_generate; when completed, downloads the MP4 to CARPE_OUTPUT_DIR (or the temp dir) and returns its path. Polls the surface that created the job. Free.

Queue music, a sound effect, or long-form speech (music-pipeline models: ElevenLabs Music, ACE-Step, Lyria, MiniMax, ElevenLabs sound effects / TTS v3-v4). Each model has its own accepted durations, lyrics and instrumental options — read them with carpe_model_details first. Returns a job_id for carpe_music_status.

Check a music job from carpe_music_generate; when completed, the track is saved to disk (it can be retrieved only once on /v1) and returned. Free.