lake_query

// WHEN AI AGENTS USE THIS TOOL

AI agents call lake_query to retrieve information from AWS Labs CloudTrail MCP Server without modifying any data. This is common in research, monitoring, and reporting workflows where the agent needs context before taking action. Because read operations don't change state, they are generally safe to allow without restrictions -- but you may still want rate limits to control API costs.

// WHY ENFORCE A POLICY ON LAKE_QUERY

Even though lake_query only reads data, uncontrolled read access can leak sensitive information or rack up API costs. An agent caught in a retry loop could make thousands of calls per minute. A rate limit gives you a safety net without blocking legitimate use.

// RECOMMENDED POLICY

Read-only tools are safe to allow by default. No rate limit needed unless you want to control costs.

aws-labs-cloudtrail-mcp-server.yaml

tools:
  lake_query:
    rules:
      - action: allow

See the full AWS Labs CloudTrail MCP Server policy for all 5 tools.

// DETAILS

Tool Name lake_query

Category Read

MCP Server AWS Labs CloudTrail MCP Server MCP Server

Risk Level Low

// MORE AWS LABS CLOUDTRAIL MCP SERVER TOOLS

Read get_query_results Read get_query_status Read list_event_data_stores Read lookup_events

// WHAT CAN GO WRONG

Agents calling read-class tools like lake_query have been implicated in these attack patterns. Read the full case and prevention policy for each:

Browse the full MCP Attack Database →

// OTHER READ TOOLS ACROSS MCP SERVERS

Other tools in the Read risk category across the catalogue. The same policy patterns (rate-limit, allow) apply to each.

Firecrawl Web Scraping Server firecrawl_check_crawl_status Firecrawl Web Scraping Server firecrawl_crawl Firecrawl Web Scraping Server firecrawl_deep_research Firecrawl Web Scraping Server firecrawl_extract Firecrawl Web Scraping Server firecrawl_map Firecrawl Web Scraping Server firecrawl_scrape

// RELATED READING

// FAQ

What does the lake_query tool do? +

Execute a SQL query against CloudTrail Lake for complex analytics and filtering. CloudTrail Lake allows you to run SQL queries against your CloudTrail events for advanced analysis. This is more powerful than the basic lookup functions and allows for complex filtering, aggregation, and analysis. PAGINATION WORKFLOW: For large result sets, you have two options: 1. Use wait_for_completion=False to get the query_id immediately, then use get_query_results with pagination 2. Use wait_for_completion=True (default) to get first page of results, then use get_query_results with next_token for additional pages IMPORTANT LIMITATIONS: - CloudTrail Lake only supports SELECT statements using Trino-compatible SQL syntax - INSERT, UPDATE, DELETE, CREATE, DROP, and other DDL/DML operations are not supported - Do not use Common Table Expression (CTE) - Your SQL query MUST include a valid Event Data Store (EDS) ID in the FROM clause - Use the list_event_data_stores tool first to get available EDS IDs, then reference the EDS ID directly in your FROM clause - Always use a start and end time using eventtime or have a limit on total output by default CLOUDTRAIL EVENT SCHEMA: All CloudTrail events contain these key fields that you can query: Core Fields (Always Present): - eventTime: UTC timestamp when request completed - eventVersion: Log format version (current: 1.11) - eventSource: AWS service name (e.g., "s3.amazonaws.com") - eventName: API action name - awsRegion: AWS region where request was made - sourceIPAddress: IP address of requester - eventID: Unique GUID for this event - eventType: AwsApiCall, AwsServiceEvent, AwsConsoleAction, AwsConsoleSignIn, AwsVpceEvent - eventCategory: Management, Data, NetworkActivity, Insight UserIdentity Object (Always Present): - userIdentity.type: Root, IAMUser, AssumedRole, Role, FederatedUser, Directory, AWSAccount, AWSService, IdentityCenterUser, SAMLUser, WebIdentityUser, Unknown - userIdentity.principalId: Unique identifier for the entity - userIdentity.arn: ARN of the principal - userIdentity.accountId: Account that owns the entity - userIdentity.accessKeyId: Access key used (may be empty for security) - userIdentity.userName: Friendly name (when available) - userIdentity.invokedBy: AWS service that made the request - userIdentity.identityProvider: External identity provider (SAML/Web) - userIdentity.credentialId: Bearer token credential ID - userIdentity.sessionContext: For temporary credentials (AssumedRole, FederatedUser) - sessionIssuer.type: Source type (Root, IAMUser, Role) - sessionIssuer.principalId: Internal ID of issuer - sessionIssuer.arn: ARN of issuer - sessionIssuer.accountId: Account of issuer - sessionIssuer.userName: Name of credential issuer - attributes.mfaAuthenticated: "true"/"false" if MFA was used - attributes.creationDate: When credentials were issued (ISO 8601) - webIdFederationData.federatedProvider: Identity provider name - webIdFederationData.attributes: Provider-specific attributes - sourceIdentity: Original user identity for role chaining - ec2RoleDelivery: "1.0" or "2.0" for IMDS version - assumedRoot: True for AssumeRoot sessions - userIdentity.onBehalfOf: IAM Identity Center user info - userId: Identity Center user ID - identityStoreArn: Identity store ARN - userIdentity.inScopeOf: Service scope information - sourceArn: Invoking resource ARN - sourceAccount: Source account ID - issuerType: Credential issuer type - credentialsIssuedTo: Credential target resource Optional Fields (Conditionally Present): - userAgent: Client that made the request (max 1KB) - errorCode: AWS service error code if request failed (max 1KB) - errorMessage: Error description if request failed (max 1KB) - requestParameters: Request parameters (object, max 100KB) - responseElements: Response elements for write operations (object, max 100KB) - additionalEventData: Additional event data (object, max 28KB) - requestID: Service-generated request identifier (max 1KB) - apiVersion: API version for AwsApiCall events - managementEvent: True if management event - readOnly: true/false if read-only operation - resources: Array of resources accessed - resources[].type: Resource type (e.g., "AWS::S3::Object", "AWS::DynamoDB::Table") - resources[].ARN: Resource ARN - resources[].accountId: Resource owner account - recipientAccountId: Account that received the event - serviceEventDetails: Service event details (object, max 100KB) - sharedEventID: Shared GUID for cross-account events - vpcEndpointId: VPC endpoint identifier (for network events) - vpcEndpointAccountId: VPC endpoint owner account - addendum: Information about delayed/updated events - reason: Why event was delayed (DELIVERY_DELAY, UPDATED_DATA, SERVICE_OUTAGE) - updatedFields: Event record fields updated by addendum - originalRequestID: Original unique ID of request - originalEventID: Original event ID - sessionCredentialFromConsole: "true" if from console session - eventContext: Enriched event context (tags, IAM conditions) - requestContext: IAM condition keys evaluated during authorization - tagContext: Tags associated with resources and IAM principals - resourceTags: Array of resource tag information - resourceTags[].arn: ARN of the tagged resource - resourceTags[].tags: Object containing tag key-value pairs - principalTags: Tags associated with the IAM principal making the request - edgeDeviceDetails: Edge device information (object, max 28KB) - tlsDetails: TLS connection information - tlsVersion: TLS version used - cipherSuite: Cipher suite used - clientProvidedHostHeader: Client-provided hostname Example SQL queries: - SELECT eventname, count(*) FROM eds-id WHERE eventtime > '2025-01-01 00:00:00' GROUP BY eventname - SELECT errorcode, errormessage, eventname FROM eds-id WHERE errorcode IS NOT NULL OR errormessage IS NOT NULL LIMIT 10 - SELECT eventname, resources FROM eds-id WHERE any_match(resources, x -> x.type = 'AWS::S3::Object') LIMIT 10 - SELECT useridentity.sessioncontext.sessionissuer.username FROM eds-id WHERE useridentity.type = 'AssumedRole' LIMIT 10 - SELECT sourceipaddress, count(*) FROM eds-id WHERE eventname = 'ConsoleLogin' GROUP BY sourceipaddress LIMIT 10 - SELECT eventname, filter(resources, x -> x.type = 'AWS::Lambda::Function') as lambda_resources FROM eds-id WHERE cardinality(filter(resources, x -> x.type = 'AWS::Lambda::Function')) > 0 LIMIT 5 Returns: -------- QueryResult containing: - query_id: Unique identifier for the query - query_status: Current status of the query - query_result_rows: Results if query completed successfully (only when wait_for_completion=True) - next_token: Token for pagination (only when wait_for_completion=True and results are paginated) - query_statistics: Performance statistics for the query. It is categorised as a Read tool in the AWS Labs CloudTrail MCP Server MCP Server, which means it retrieves data without modifying state.

How do I enforce a policy on lake_query? +

Add a rule in your Intercept YAML policy under the tools section for lake_query. You can allow, deny, rate-limit, or validate arguments. Then run Intercept as a proxy in front of the AWS Labs CloudTrail MCP Server MCP server.

What risk level is lake_query? +

lake_query is a Read tool with low risk. Read-only tools are generally safe to allow by default.

Can I rate-limit lake_query? +

Yes. Add a rate_limit block to the lake_query rule in your Intercept policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.

How do I block lake_query completely? +

Set action: deny in the Intercept policy for lake_query. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.

What MCP server provides lake_query? +

lake_query is provided by the AWS Labs CloudTrail MCP Server MCP server (awslabs.cloudtrail-mcp-server). Intercept sits as a proxy in front of this server to enforce policies before tool calls reach the server.

Let agents act without letting them run wild.