MongoDB Query Optimizer

When this skill is invoked

Invoke only when the user wants:

• Query/index optimization or performance help
• Why a query is slow or how to speed it up
• Slow queries on their cluster and/or how to optimize them

Do not invoke for routine query authoring unless the user has requested help with optimization, slow queries, or indexing.

High Level Workflow

General Performance Help

If the user wants to examine slow queries, or is looking for general performance suggestions (not regarding any particular query):

• Use MongoDB MCP server atlas-get-performance-advisor tool to fetch slow query logs and performance advisor output
• Make suggestions based on this information

If Atlas MCP Server for Atlas is not configured or you don’t have enough information to run atlas-get-performance-advisor against the correct cluster, tell the user that general performance analysis requires Atlas MCP Server configuration with API credentials, and suggest they configure it or ask about a specific query instead.

Help with a Specific Query

If the user is asking about a particular query:

• Use collection-indexes, explain, and find MCP tools to get existing indexes on the collection, explain() output for the query, and a sample document from the collection
• Use atlas-get-performance-advisor MCP tool to fetch slow query logs and performance advisor output

Then make an optimization suggestion based on collected information and MongoDB best practices and examples from reference files. Prefer creating an index that fully covers the query if possible. If you cannot use MongoDB MCP Server then still try to make a suggestion.

MCP: available tools

How to invoke. Call the MongoDB MCP server with the exact tool name as toolName and a single arguments object as arguments. Do not pass the tool name as an option, query param, or nested key; pass it as the MCP tool name and the parameters as the arguments object. Full MCP Server tool reference: MongoDB MCP Server Tools (opens in a new tab).

Database tools (when the MCP cluster connection works):

Atlas tools (when Atlas API credentials are configured):

For a user question, try to fetch information from both the connection string and Atlas API related to the query you are optimizing.

1. DB connection string works for MongoDB MCP

Typical flow: call collection-indexes → explain → find (sample doc).

• collection-indexes — Use the result’s classicIndexes (each has name, key) to see if the query can already use an existing index.
• explain — Run in "queryPlanner" mode first to check for COLLSCAN. If the query uses an index or the collection is very small, run again with "executionStats" (10-second timeout) to get docs scanned vs. returned.

2. Atlas API access works for MongoDB MCP

If you need a project ID, call atlas-list-projects first. Then call atlas-get-performance-advisor with only the operations you need:

Do not pass the MCP tool name as an operations value—operations is a separate argument listing what data to fetch.

Example workflow 1 (help with specific query)

User: “Why is this query slow? db.orders.find({status: 'shipped', region: 'US'}).sort({date: -1})“

If MCP db connection is configured and the database + collection names are known, run steps 1–3. Otherwise skip to step 4.

1. Check existing collection indexes:

   • Call collection-indexes with database=store, collection=orders
   • Result shows: {_id: 1}, {status: 1}, {date: -1}

2. Run explain:

   • Call explain with method=find, filter={status: 'shipped', region: 'US'}, sort={date: -1}, verbosity=queryPlanner and executionStats
   • Result: Uses {status: 1} index, then in-memory SORT, totalKeysExamined: 50000, nReturned: 100

3. Run find:

   • Call find with limit=1 to fetch a sample document to impute the schema.

If MCP Atlas connection is configured, run step 4. Otherwise skip to step 5.

4. Run atlas-get-performance-advisor:

   • Try to get the cluster name from the MCP connection string, or ask the user for projectId/clusterName
   • Use slowQueryLogs to fetch slow query logs from database=store, collection=orders in the past 24 hours
   • Use suggestedIndexes to check for index suggestions for the query

5. Diagnose: Based on explain output and slow query logs, this query targets 100 docs but scans 50K index entries (poor selectivity: 0.002). In-memory sort adds overhead. Index doesn’t support both filter fields or sort.

6. Recommend: Create compound index {status: 1, region: 1, date: -1} following ESR (two equality fields, then sort). This eliminates in-memory sort and improves selectivity by filtering on both status and region.

If the MongoDB MCP server is not set up, follow best indexing practices.

Example workflow 2 (general database performance help)

User: “Can you help with optimizing slow queries on my cluster?”

1. Run atlas-get-performance-advisor:
   • Try to get the cluster name from the connection string and deduce the project name you need in atlas-list-projects; if you are not sure, then ask the user for cluster name and project id.
   • Use slowQueryLogs to fetch slow query logs from the past 24 hours
   • Use suggestedIndexes
   • Use dropIndexSuggestions
   • Use schemaSuggestions
2. Diagnose and Recommend: Based on slow query logs and performance advisor advice, you can create the compound index {status: 1, region: 1, date: -1} on the db.orders collection to optimize queries such as find({status: 'shipped', region: 'US'}).sort({date: -1})

Examine all performance advisor output as well as slow query logs. Provide information on what is being improved and why, and focus on suggestions that have the potential for greatest impact (e.g., indexes that affect the most queries, or queries that have the worst performance).

Load references

Before beginning diagnosis and recommendation, load reference files.

Always load:

• references/core-indexing-principles.md
• references/antipattern-examples.md

Conditionally load these files:

• If diagnosing aggregation pipelines → references/aggregation-optimization.md
• If diagnosing queries that change docs such as replaceOne, findOneAndUpdate, etc. → references/update-query-examples.md for oplog-efficient updates and common update anti-patterns

Output

• Keep answers short and clear: a few sentences on index and optimization suggestions, and reasoning behind them (e.g. general indexing principles, observing slow query logs in the cluster, or seeing advice in Performance Advisor)
• Focus on highest impact indexes or optimizations - if you’ve omitted some optimizations let the user know and present them if asked.
• Do not use strong language, such as saying “You should create these indexes and they will definitely improve application performance” - Explain they are suggestions for certain queries, and give the reasoning behind them.
• Consider how many indexes already exist on the collection (if known) - there shouldn’t generally be more than 20
• Suggest removing indexes only if the suggestion comes from Atlas Performance Advisor
• Do not create indexes directly via MCP unless the user gives approval