Summary
An ArangoDB session that survives being asked vague questions.
Describing the database and sampling documents come before querying, so a model that has never seen your schema can find it rather than invent it. The read-only query path and the confirm flag on dropping a collection are where that exploration stops short of damage.
What it is
A full working surface for one ArangoDB database: AQL execution with a read-only mode, document CRUD in bulk, collection and index management, graph traversal, and ArangoSearch views and analyzers.
What you get
- AQL with bind variables and bounded results, plus a read-only variant that rejects writes and DDL
- Query validation and execution plans, so a query can be checked before it runs
- Database discovery — collections, counts, indexes and sample fields in one summary
- Documents by key, listed with limit and offset, counted, or randomly sampled to learn a schema
- Writes one at a time, or up to 1000 documents per bulk insert or bulk update
- Collections created and dropped, the drop requiring an explicit confirm
- Indexes: persistent, geo, TTL and inverted
- Graphs, edges, traversal and shortest path
- ArangoSearch views, analyzers and search, plus a database backup
Requirements
A reachable ArangoDB instance and credentials for the database you point it at. It runs as a TypeScript server launched from your MCP client config.
Setup effort
One command plus a key — npm install -g arango-server, then supply credentials