Download one file
then check it
One self-contained binary: no package manager, no runtime, nothing else to install.
- Check the digest The SHA-256 your machine prints matches the one published with this release.
- Signed, with SBOMs Every artifact is cosign-signed and ships with SPDX and CycloneDX SBOMs. The full check
curl -sSO https://dl.cameodb.com/mac/cameodb shasum -a 256 ./cameodb chmod +x ./cameodb
286ffa398dd67d974df7734f41e720fc0ddea3a5d935b7d4bbd3eb0d3d78ea88curl -sSO https://dl.cameodb.com/linux/cameodb sha256sum ./cameodb chmod +x ./cameodb
11cb60fe6fd1499017ba6701c0604c252858ee91c5b32bd02592779e7754d1e9curl.exe -sSO https://dl.cameodb.com/windows/cameodb.exe
Get-FileHash .\cameodb.exe -Algorithm SHA256
# the .exe runs as downloaded
37f3aeb4745a1c0d4a5135d78ddcf4cdc5f393c6be7c818aa6cb97de85536fb7Run it
It starts in the foreground with sensible defaults and no config file. Storage is created under data/cameodb in the directory you start it from.
-
HTTP and MCP on one port
The API on
127.0.0.1:9480, and the MCP endpoint at/mcpbeside it, six tools ready. - Its posture, checked at startup The node reads how far it can be reached and checks every setting against it before it serves.
-
Settings when you want them
generate-configwrites every setting at its default tocameodb.toml, andcheck-configvalidates it.
./cameodb.\cameodb.exeInitialized 4 shards 🚀 CameoDB HTTP Server starting on http://127.0.0.1:9480 MCP protocol: 2025-06-18 (negotiated), capabilities: tools (6), resources (4), prompts (1) … 🔒 Security profile: local (inferred from a loopback bind address; declare it with `profile` under [node]) [PASS] bind 127.0.0.1 (loopback only) [PASS] cors no cross-origin browser access [PASS] node_key node_identity.json is 0600 (owner only) [PASS] auth unauthenticated (loopback only) … Press Ctrl+C to shutdown...
Connect the client
and check it is alive
Open a second terminal and start the interactive client. It ships inside the same binary and connects to localhost:9480 on its own.
- A shell that knows your data Auto-completion, history and colorized JSON, with your indexes and fields one tab away.
-
Green from the first command
healthanswers for the node and its shards;list indexesshows what it holds.
./cameodb client -i.\cameodb.exe client --interactivecameodb@localhost ▶ health { "status": "green", "node_id": "c2391921-30da-5f66-9068-73d1bec79ef1", "active_shards": 4, "cluster_name": "cameodb-cluster", "cluster_enabled": false, "total_nodes": 1, "connected_nodes": 1 }
Ingest a dataset
One command downloads 16,559 book summaries, detects the schema and loads them. There is no schema to write by hand first.
- Schema inferred This file names its column types in its header; any other file is typed from the values each column holds.
- Batched for speed Rows go in batches of 4,000 documents. CSV, TSV, JSON and JSONL, gzip or zip, from disk or a URL.
- Searchable on commit Nothing to reindex afterwards: every document is searchable after the next commit.
cameodb@localhost ▶ data load books https://dl.cameodb.com/examples/data/booksummaries.tsv Schema was missing; detected and applied schema to index 'books' Ingestion complete for index 'books': loaded=16559 repeated_ids=0 skipped=0 failed=0 (batch size 4000, parallel 1)
Search it, then
ask for less
A phrase match on title finds all seven books. Add return and each hit carries only the fields you name, projected inside the engine.
The same index answers over MCP, so an agent can ask this exact question with nothing more to set up.
cameodb@localhost ▶ search books title:"Harry Potter" return title,publication_date sort publication_date limit 7 { "hits": [ { "title": "Harry Potter and the Philosopher's Stone", "publication_date": "1997-06-30" }, { "title": "Harry Potter and the Chamber of Secrets", "publication_date": "1998-07-02" }, … { "title": "Harry Potter and the Deathly Hallows", "publication_date": "2007-07-21" } ], "total_hits": 7, "took_ms": 0, "stats": { "shards": { "responded": 4, "failed": 0 } } }
You have a working
intelligence layer
The fastest next step is pointing an agent at it, and that is one command away.
# Claude Code: one line, and the agent has your data claude mcp add --transport http cameodb \ http://localhost:9480/mcp # then ask it, in plain words > Which Harry Potter books came out before 2000?