# Proxlai for AI agents

Proxlai (https://proxlai.com) tracks prices for every AI API: current prices and capabilities of 293 AI APIs from 144 providers across 45 markets, each price in
the market's own unit, with the date it was verified and a link to the provider's pricing page.

Follow the steps in order. Ask the person you work for before anything that needs their account or their money.

## 1. Connect the MCP server (preferred)

The server is read-only, needs no account, and speaks Streamable HTTP at:

    https://proxlai.com/api/mcp

Claude Code:

    claude mcp add --transport http proxlai https://proxlai.com/api/mcp

Clients configured with a JSON file:

    { "mcpServers": { "proxlai": { "url": "https://proxlai.com/api/mcp" } } }

Clients that only run local (stdio) servers, through the mcp-remote bridge:

    { "mcpServers": { "proxlai": { "command": "npx", "args": ["-y", "mcp-remote", "https://proxlai.com/api/mcp"] } } }

Tools:
- `list_markets`: every market with its id and pricing unit. Start here.
- `describe_market`: what a market is, how its pricing works, and the attributes you can filter on.
- `search_offerings`: rank a market's offerings by price, quality and speed, with requirements and a maximum price.
- `get_offering`: one offering in full: capabilities, current price with its source, and price history.
- `compare_offerings`: 2 to 4 offerings from one market side by side.

Without a key: 30 calls a minute and 100 searches or lookups a day per address.

## 2. Add an API key (optional, for more requests)

1. The person signs in at https://proxlai.com/login and creates a key at https://proxlai.com/dashboard. Keys start with `aac_live_`.
2. Store it as the environment variable `PROXLAI_API_KEY`. Never paste it into code, commits, logs or chat.
3. Send it as `Authorization: Bearer <key>`, for MCP too:

       claude mcp add --transport http proxlai https://proxlai.com/api/mcp --header "Authorization: Bearer $PROXLAI_API_KEY"

Each account gets 100 free searches or lookups a day across all its keys, then $0.01 each from its
credit balance. Rate limit: 60 requests a minute per key.

## 3. Talk to the REST API

`POST https://proxlai.com/api/v1/search` ranks one market's offerings (one billable request):

    curl -X POST https://proxlai.com/api/v1/search \
      -H "Authorization: Bearer $PROXLAI_API_KEY" \
      -H "Content-Type: application/json" \
      -d '{"category": "llm", "limit": 5}'

Body fields:
- `category` (required): a market id from the list below.
- `filters`: `[{ "attr", "op", "value" }]`, op is eq, lte, gte, in or has; attr is one of the market's attributes.
- `maxPrice`: upper bound in the market's unit (USD).
- `weights`: `{ "cost", "quality", "speed" }`, each 0 to 1.
- `includeUnverified`: true to also rank unverified sample prices (left out by default).
- `sort`: score, price or quality. `limit`: 1 to 50.

`GET https://proxlai.com/api/v1/offerings` lists the markets and their attributes (free); add `?category=<id>` for every offering and
price in one market (one billable request).

Errors: 400 invalid_query, 401 bad key, 402 insufficient_credits, 429 rate limited.

## Markets

- `llm`: Large language models (priced USD per 1M output tokens)
- `embeddings`: Embeddings (priced USD per 1M input tokens)
- `reranking`: Reranking (priced USD per 1K searches)
- `translation`: Machine translation (priced USD per 1M characters)
- `image-generation`: Image generation (priced USD per image (≈1 megapixel))
- `image-editing`: Image editing & upscaling (priced USD per image)
- `video-generation`: Video generation (priced USD per second of video)
- `avatars-lipsync`: Avatars & lip-sync (priced USD per minute of video)
- `3d-generation`: 3D model generation (priced USD per generated model)
- `text-to-speech`: Text-to-speech & voice cloning (priced USD per 1M characters)
- `speech-to-text`: Speech-to-text (priced USD per hour of audio)
- `music-audio`: Music & sound generation (priced USD per minute of audio)
- `gpu-on-demand`: GPU cloud (on-demand) (priced USD per GPU-hour)
- `gpu-marketplace`: GPU marketplace & spot (priced USD per GPU-hour)
- `gpu-serverless`: Serverless GPU (priced USD per GPU-hour (billed per second))
- `inference-hosting`: Hosted open-model inference (priced USD per 1M output tokens)
- `fine-tuning`: Fine-tuning (priced USD per 1M training tokens)
- `vector-db`: Vector databases (priced USD per month for 1M vectors (1024-dim))
- `ocr-document`: OCR & document AI (priced USD per 1K pages)
- `web-search`: Web search & grounding APIs (priced USD per 1K queries)
- `browser-agents`: Browser & agent infrastructure (priced USD per browser-hour)
- `moderation`: Moderation & safety (priced USD per 1K requests)
- `speech-to-speech`: Speech-to-speech & voice changing (priced USD per minute of audio)
- `ai-dubbing`: AI dubbing & speech translation (priced USD per minute of source video)
- `sound-effects`: Sound-effects generation (priced USD per minute of audio)
- `audio-enhancement`: Audio enhancement & stem separation (priced USD per minute of audio)
- `video-enhancement`: Video enhancement & upscaling (priced USD per minute of input video)
- `video-background-removal`: Video background removal (priced USD per minute of video)
- `motion-capture`: Motion capture & text-to-motion (priced USD per minute of animation)
- `texture-generation`: Texture & material generation (priced USD per texture set)
- `game-asset-generation`: Game asset & sprite generation (priced USD per asset)
- `vision-language`: Vision-language & image understanding (priced USD per 1K images (≈1 MP, input only))
- `code-generation`: Code generation & completion (priced USD per 1M output tokens)
- `ai-accelerators`: TPUs, AI chips & CPU endpoints (priced USD per accelerator- or instance-hour)
- `llm-gateway`: LLM gateways & routers (priced USD per month at 1M routed requests (≈ $5,000 of model spend))
- `llm-observability`: LLM observability & evals (priced USD per month at 1M traced events)
- `hosted-agents`: Hosted agent platforms (priced USD per agent-hour (active runtime))
- `recommendations`: Recommendation APIs (priced USD per 1K recommendation requests)
- `voice-agents`: Voice agents (priced USD per minute of conversation)
- `speech-analytics`: Speech analytics & call intelligence (priced USD per audio hour)
- `computer-vision`: Computer vision APIs (priced USD per 1K images)
- `web-scraping`: Web scraping for AI (priced USD per 1K pages)
- `rag-memory`: RAG & memory services (priced USD per 1K queries)
- `pii-redaction`: PII detection & redaction (priced USD per 1M characters)
- `ai-text-detection`: AI text detection (priced USD per 1K words)

## Rules for quoting prices

- Always give the unit and the verified date with a price, and link the provider's `source_url`.
- Say so when a price is an unverified sample.
- Prices change often: tell the person to confirm on the provider's site before buying.

Human docs: https://proxlai.com/docs/mcp and https://proxlai.com/docs/api.