CameoDB embeds MCP directly in a Rust query engine and handles 80M writes a day
Negative_Ad5847 · reddit · 2026-07-23
A Rust database built MCP into the query engine itself
Instead of wrapping Postgres or Elasticsearch with a thin MCP server, the team embedded MCP directly into the CameoDB binary, exposing SSE and JSON-RPC transports from the actual query engine. That means agents talk to the database itself rather than to a translation layer.
Under the hood
- Hybrid storage: each shard combines an embedded ACID key-value store (redb) and full-text search (tantivy) as one atomic unit
- No leader node; shards are distributed with consistent hashing
- Recent shipping work includes jemalloc with per-shard memory budgeting and live admin stats
- Startup WAL replay is stricter so unclean shutdowns do not leave KV and search index out of sync
- There is a corruption guard on the sequence counter
Real-world test load
- Tested on a 64-core machine with 35TB NVMe
- 16 shards
- About 400 million records across 20TB+
- Ingesting roughly 80 million records/day in real time
- P99 latency is under a second for most queries, depending on query complexity
The original product pitch was about avoiding dual clusters for transactional storage and search. MCP became the attention magnet once agents started talking to the database directly.
More from coding & agent
- Citrolabs opens ego-lite, a browser for humans and AI agents to work in parallel — citrolabs · 2026-07-23
- Alibaba open-sources open-code-review, a hybrid LLM agent for line-level code review — alibaba · 2026-07-23
- Qwen3.8-Max looks stronger on 3D web frontends than Kimi K3 — cedric_chee · 2026-07-23
- A three-agent setup is helping a full-autonomy coding experiment work — MoonL88537 · 2026-07-23
- A full-autonomy build says models can now assemble entire systems from scratch — MoonL88537 · 2026-07-23
- OpenAI’s Codex Micro is mocked as a token-burning setup trap — jdjohnson · 2026-07-23