@mcpolyglot/connector-mongo

MongoDB connector for mcpolyglot. Sample-based schema inference, read-only find / aggregate primitives, prompt-injection-safe results.

Tools exposed

For every Mongo source mcpolyglot generates:

• <id>.list_collections — all collections with a sampled-field schema for each.
• <id>.describe_collection — one collection's sampled-field schema (paths, types, nullability).
• <id>.find — read-only find with filter, projection, sort, and a bounded limit.
• <id>.aggregate — read-only aggregation pipeline; mutating stages are rejected.

How read-only is enforced

Mongo doesn't have a one-shot "read-only transaction" knob, so mcpolyglot enforces it at the API surface:

1. Only find and aggregate are exposed — no update, insert, delete, or drop primitives.
2. Aggregation pipelines are walked before they reach the driver. Any stage with $out or $merge (the two stages that write back) is rejected with forbidden.read_only.
3. Every operation is bounded by maxTimeMS so a runaway aggregation can't pin the cluster.
4. Schema is inferred from a 25-doc $sample per collection — no full-collection scans on list_collections.

Driver is an optional dep

mongodb is declared as an optional dep. Install it once with your project and you're set.

Docs

• Architecture → https://github.com/ishay60/mcpolyglot/blob/develop/ARCHITECTURE.md
• Example → examples/mongo

MIT licensed.