# AImart Parquet > Converts messy CSV, JSON, Excel, SPSS, Stata and SAS files, and tables in digital (not scanned) PDFs, into clean, typed, schema-validated, zstd-compressed Parquet — then inspects, previews and queries it with DuckDB SQL, including joins across converted files. Built for AI agents: no signup, pay per call in USDC via x402. Predictable cost: a fixed price per call, so a spend cap is exact, and a call that fails is not charged. Every paid call returns a receipt naming its on-chain transaction, and hosted outputs carry a signed (ML-DSA-65) provenance record. Every paid call answers HTTP 402 first with a base64 `PAYMENT-REQUIRED` header (x402 v2). Sign it with an x402 client and retry with `PAYMENT-SIGNATURE`. You are charged only when the call succeeds. ## Use it - [MCP server (Streamable HTTP)](https://balogvn--aimart-parquet.us-east.modal.run/mcp): tools convert_to_parquet, inspect_parquet, preview_parquet, query_parquet - [OpenAPI](https://balogvn--aimart-parquet.us-east.modal.run/openapi.json): REST endpoints under /v1 - [Pricing](https://balogvn--aimart-parquet.us-east.modal.run/pricing): per-call prices in USDC - [x402 discovery](https://balogvn--aimart-parquet.us-east.modal.run/.well-known/x402): payable resources ## Operations - `POST /v1/convert` ($0.02): Convert a messy CSV/JSON/Excel/SPSS/Stata/SAS file, or tables in a digital (not scanned) PDF, to clean Parquet (input by URL or base64). - `POST /v1/convert/upload` ($0.02): Convert an uploaded (multipart) CSV/JSON/Excel/SPSS/Stata/SAS file, or tables in a digital (not scanned) PDF, to clean Parquet. - `POST /v1/inspect` ($0.002): Describe a Parquet file's schema, null counts, ranges and labels without reading rows. - `POST /v1/preview` ($0.002): Return a slice of Parquet rows as JSON, Markdown or CSV. - `POST /v1/query` ($0.005): Run one read-only DuckDB SELECT over a Parquet file (table `data`); join up to 7 more artifacts via `tables`. ## Notes - Inputs: `url` (preferred), `content_base64` + `filename`, or `artifact_id` from an earlier convert. - Over MCP each tool takes only its own inputs: `convert_to_parquet` takes `url`, `content_base64` or `artifact_id`; `inspect_parquet` takes `url` or `artifact_id`; `preview_parquet` takes `url` or `artifact_id`; `query_parquet` takes `url` or `artifact_id`. On REST, convert, inspect, preview and query take all three. - Data handling: inputs are deleted when the call ends; converted files become unreachable 60 minutes after creation and are deleted after that by a cleanup that runs when the service starts and once a minute while it runs; a host that stops the service when idle (the hosted one does) deletes a file that expired meanwhile when it next starts, which can be hours later. File contents are never logged or used for training. - Missing something? Call the free `request_capability` tool (or POST /v1/feedback) with the task and what you would pay. - convert returns an `artifact_id` and a signed `download_url`; artifacts expire after 60 minutes. - Supported extensions: .csv, .dta, .geojson, .json, .jsonl, .ndjson, .ods, .parquet, .pdf, .por, .pq, .sas7bdat, .sav, .tab, .tsv, .txt, .xls, .xlsb, .xlsm, .xlsx, .xpt, .zsav (optionally .gz/.bz2/.zip). - query accepts one read-only SELECT. Its input is table `data`; `tables` joins up to 7 more artifacts by name (8 tables in total). REST: `"tables": {"orders": {"artifact_id": "..."}}`; MCP: `"tables": {"orders": ""}`. ## Optional - [Source code](https://github.com/balogvn/aimart): Apache-2.0; run it locally for free as a stdio MCP server