Vesper, launched via a Show HN submission, is an MCP server that lets AI agents manage ML dataset preparation from discovery through export. It pulls from Kaggle and Hugging Face, runs quality scoring, deduplication, compliance checks, cleaning, and merging, then splits and exports to CSV, Parquet, Arrow, or JSONL. No UI, no human in the loop. The developers say a first complete workflow runs in under ten minutes. Pricing is listed as free for a limited time, and the contact is a personal Gmail address — both signs this is a project-stage release rather than a commercial product.
What Vesper offers over a standard dataset library wrapped in an API is direct <a href="/news/2026-03-14-ink-agent-native-infrastructure-platform-mcp">MCP compatibility</a>. Any MCP-compatible agent or orchestration framework can invoke its operations as first-class tool calls without custom glue code. The developers describe outputs as deterministic, inspectable, and reproducible, saying this was a direct design goal given how hard non-deterministic agentic pipelines are to debug. That framing is plausible, though it's the project's own claim — independent validation doesn't exist yet.
Stated use cases include <a href="/news/2026-03-14-agentlog-lightweight-kafka-like-event-bus-for-ai-agent-orchestration-via-jsonl">agent-driven experimentation loops</a>, RAG corpus construction, vision and audio model training, and healthcare datasets where compliance documentation matters. Local and private deployment are supported for data residency requirements.
MCP tooling for data infrastructure remains thin. Most teams today write their own connectors or route through general-purpose frameworks like LangChain's data loaders. Vesper is one of the few projects targeting dataset workflows specifically as an MCP surface. Whether that narrow focus builds an audience or limits it depends on how fast MCP adoption spreads beyond early adopters — and right now that adoption is still concentrated in a small slice of teams building agent pipelines from scratch.