mongo-db（mongo DB 搜索）

MongoDB — Document Storage Skill

Use this skill whenever an agent needs to read or write persistent data to MongoDB. All operations are performed via a Python CLI script and return JSON output.

When to use

• User asks to "save to the database", "store this in Mongo", "retrieve from MongoDB"
• An agent needs to persist data across sessions (budgets, transactions, summaries, watchlists)
• An agent needs to query, filter, or aggregate stored records
• A new collection or schema (validator) needs to be set up

Setup (first run only)

Local MongoDB: If you need to install MongoDB on Ubuntu, see INSTALL-UBUNTU.md. For Atlas, use your connection string in config or env.

Run the setup script once from the workspace root to create a virtual environment and install pymongo:

bash skills/mongo-db/scripts/setup.sh

This creates skills/mongo-db/scripts/.venv/ and installs dependencies there. After setup, invoke the client using the venv interpreter:

skills/mongo-db/scripts/.venv/bin/python3 skills/mongo-db/scripts/mongo_client.py '<json>'

Configuration

Connection is resolved in this order (first match wins):

Option 1 — Environment variable (recommended)

MONGO_URI=mongodb+srv://user:password@cluster.mongodb.net/mydb
MONGO_DB=mydb   # optional if the db is in the URI

Option 2 — config.json (local file, gitignored)

Copy the example and fill in your values:

cp skills/mongo-db/config.example.json skills/mongo-db/config.json

Edit skills/mongo-db/config.json:

{
  "uri": "mongodb://localhost:27017",
  "database": "mydb",
  "username": "optional",
  "password": "optional"
}

Option 3 — Individual env vars

MONGO_HOST=localhost
MONGO_PORT=27017
MONGO_USER=myuser
MONGO_PASSWORD=mypassword
MONGO_DB=mydb

CLI interface

All operations use a single JSON payload argument:

PYTHON=skills/mongo-db/scripts/.venv/bin/python3
$PYTHON skills/mongo-db/scripts/mongo_client.py '<json_payload>'

Payload schema:

{
  "operation": "<op>",
  "database": "<optional override>",
  "collection": "<collection name>",
  ...operation-specific fields
}

Output is always JSON: {"success": true, ...result_fields} or {"success": false, "error": "..."}.

Operations Reference

List databases

{"operation": "list_databases"}

List collections in a database

{"operation": "list_collections", "database": "mydb"}

Create a collection (with optional JSON Schema validator)

{
  "operation": "create_collection",
  "database": "mydb",
  "collection": "budgets",
  "validator": {
    "$jsonSchema": {
      "bsonType": "object",
      "required": ["period", "income"],
      "properties": {
        "period": {"bsonType": "string"},
        "income": {"bsonType": "number"}
      }
    }
  }
}

Drop a collection

Always confirm with the user before dropping. Requires "confirm": true.

{"operation": "drop_collection", "database": "mydb", "collection": "old_data", "confirm": true}

Create an index

{
  "operation": "create_index",
  "database": "mydb",
  "collection": "transactions",
  "keys": {"date": 1, "category": 1},
  "unique": false
}

Find documents

{
  "operation": "find",
  "database": "mydb",
  "collection": "transactions",
  "filter": {"category": "groceries"},
  "projection": {"_id": 0, "date": 1, "amount": 1, "description": 1},
  "sort": {"date": -1},
  "limit": 20
}

Find one document

{
  "operation": "find_one",
  "database": "mydb",
  "collection": "transactions",
  "filter": {"id": "txn_20260131_0001"}
}

Count documents

{
  "operation": "count",
  "database": "mydb",
  "collection": "transactions",
  "filter": {"type": "expense"}
}

Insert one document

{
  "operation": "insert_one",
  "database": "mydb",
  "collection": "transactions",
  "document": {
    "id": "txn_20260201_0001",
    "date": "2026-02-01",
    "description": "Supermarket",
    "amount": -85.50,
    "category": "groceries",
    "type": "expense"
  }
}

Insert many documents

{
  "operation": "insert_many",
  "database": "mydb",
  "collection": "transactions",
  "documents": [
    {"date": "2026-02-01", "description": "Coffee", "amount": -4.50, "category": "dining"},
    {"date": "2026-02-02", "description": "Salary", "amount": 5000, "category": "income"}
  ]
}

Update one document

{
  "operation": "update_one",
  "database": "mydb",
  "collection": "budget",
  "filter": {"period": "monthly"},
  "update": {"$set": {"income": 5500, "last_updated": "2026-02-01"}},
  "upsert": false
}

Update many documents

{
  "operation": "update_many",
  "database": "mydb",
  "collection": "transactions",
  "filter": {"category": "food"},
  "update": {"$set": {"category": "groceries"}}
}

Replace one document

{
  "operation": "replace_one",
  "database": "mydb",
  "collection": "budget",
  "filter": {"period": "monthly"},
  "replacement": {"period": "monthly", "income": 5500, "currency": "USD"},
  "upsert": true
}

Delete one document

Always confirm with the user before deleting. Requires "confirm": true.

{
  "operation": "delete_one",
  "database": "mydb",
  "collection": "transactions",
  "filter": {"id": "txn_20260131_0001"},
  "confirm": true
}

Delete many documents

Always confirm with the user before deleting. Requires "confirm": true.

{
  "operation": "delete_many",
  "database": "mydb",
  "collection": "transactions",
  "filter": {"source_file": "january_statement.pdf"},
  "confirm": true
}

Aggregate

{
  "operation": "aggregate",
  "database": "mydb",
  "collection": "transactions",
  "pipeline": [
    {"$match": {"type": "expense"}},
    {"$group": {"_id": "$category", "total": {"$sum": "$amount"}, "count": {"$sum": 1}}},
    {"$sort": {"total": 1}}
  ]
}

Notes

• Destructive operations (delete_one, delete_many, drop_collection) require "confirm": true in the payload. Always ask the user to confirm before including this flag.
• ObjectId fields are serialized as strings in all output.
• Database override: the "database" field in the payload overrides the default database from config/env for that single call.
• Schema validation: use create_collection with a "validator" to enforce document structure at the MongoDB level. This is the recommended approach for agents that write structured data (e.g. the finance-budget agent's budget and transaction collections).
• Upsert pattern: for config-like documents (one per type), use replace_one with "upsert": true to create-or-replace atomically.
• If pymongo is missing, re-run bash skills/mongo-db/scripts/setup.sh.