MongoDB (`mongo`)¶

The mongo connector indexes documents from a single MongoDB database. Each document becomes a searchable record, and each collection gets a sampled schema preview (Mongo has no fixed schema, so it's inferred from a sample).

How MFS sees it¶

Collections sit under the alias; each exposes its documents and a sampled schema:

mongo://prod-cluster/
└── support_threads/
    ├── documents.jsonl   record_collection  → one searchable chunk per document
    └── schema.json       table_schema       → sampled field summary

Documents are chunked per-document and need text_fields to become searchable.

Credentials¶

A MongoDB connection URI, in either form:

mongodb://user:pass@host:27017/?authSource=admin
mongodb+srv://user:pass@cluster.mongodb.net/?retryWrites=true

For Atlas, copy the SRV URI from Database → Connect → Drivers and substitute the real password. A read-only user is enough — the connector only runs find(). Probe before MFS sees it:

mongosh "$MONGO_URI" --eval "db.adminCommand({ping: 1})"

Configuration¶

uri = "env:MONGO_URI"
database = "prod"
cursor_field = "updatedAt"     # or _id; enables incremental re-sync
max_read_docs = 100000

[[objects]]
match = "/support_threads"
text_fields = ["title", "messages[].body"]
locator_fields = ["_id"]
metadata_fields = ["status"]

text_fields supports nested paths like messages[].body to pull text out of arrays of subdocuments.

Sync and freshness¶

With cursor_field set (updatedAt or _id), re-syncs pull only documents changed since the last run; deletions are caught by full_scan.

Search and browse¶

mfs connector probe mongo://prod-cluster --config ./mongo.toml
mfs add mongo://prod-cluster --config ./mongo.toml

mfs search "refund escalation" mongo://prod-cluster/support_threads/documents.jsonl
mfs cat mongo://prod-cluster/support_threads/documents.jsonl --locator '{"_id":"65a3..."}'

Pitfalls¶

Documents are heterogeneous; fields absent from a given document are simply skipped when rendering its text.
_id locators use the serialized string form, not ObjectId(...).
max_read_docs caps large collections and can mark recall partial.

MongoDB (mongo)¶