MCP server for Chrome Enterprise Premium management, providing tools for device management, DLP rules, security insights, connectors, and organizational access control.
Chrome Enterprise Premium MCP presents a comprehensive suite of 31 tools with mixed quality. Strengths: detailed parameter documentation, consistent schema structure, clear verb-based naming conventions (get_, list_, create_, check_, enable_, delete_), and explicit risk classification (READ_ONLY, WRITE, DESTRUCTIVE). Weaknesses: several tools lack output schema documentation; some descriptions are generic or repetitive; parameter relationships underdocumented; error handling guidance sparse; no tool annotations (readOnlyHint/destructiveHint) visible in MCP protocol layer despite risk metadata; missing recovery guidance in error scenarios. The server demonstrates solid foundational discipline (all tools named, described, and schematized) but falls short of production-grade polish expected at 75+. Average tool description length ~180 chars (within baseline 194), but 8 tools have thin or generic descriptions.
Authentication bootstrapper tool for managing OAuth credentials.
Checks whether Chrome Enterprise Premium APIs are enabled and enables them if needed.
Checks whether Chrome Enterprise Premium is enabled for the customer.
Checks whether the Enterprise Vault extension is installed for a user or organizational unit.
Checks whether the Safe Exam Browser (SEB) extension is installed for a user or organizational unit.
Checks whether a specific user has a Chrome Enterprise Premium license.
Counts Chrome browser versions in use by devices in the customer's organization.
Dual tool definitions for overlapping functionality (security_insights vs security_insights_data) causes LLM disambiguation overhead. Both expose AIP-160 filtering on content transfers and URL visits, naming and parameter overlap create tool selection ambiguity.
Tool 'check_and_enable_cep_api' uses 'and' in name, signaling dual responsibility (check + enable). Should be split into separate tools per pattern:tool. Creates confusion about whether both actions always run or conditionally.
Destructive tools (delete_agent_dlp_rule, delete_detector) lack confirmation/dry-run capability. No evidence of confirmation-request pattern. Agents can irreversibly delete DLP rules and detectors without reversibility or guard rails.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2025-06-18+ | v2 |
Creates a new Context-Aware Access (CAA) access level.
Creates a new Data Loss Prevention (DLP) rule in the customer's organization.
Creates default Data Loss Prevention rules for common security scenarios.
Creates a new regex-based detector for content matching in DLP rules.
Creates a new URL list detector for identifying specific domains/URLs in DLP rules.
Creates a new word list detector for identifying keywords in DLP rules.
Deletes a Data Loss Prevention (DLP) rule created by the agent.
Deletes a detector (regex, URL list, or word list) from the customer's organization.
Performs diagnostic checks on the environment and configuration for troubleshooting.
Enables Chrome Enterprise Connectors policies for content analysis.
Retrieves Chrome management activity logs for devices and users.
Retrieves Chrome Enterprise Connectors policies for a specific organizational unit.
Retrieves the Chrome customer ID for the authenticated user's Workspace account.
Retrieves detailed information about a specific DLP rule by name.
Installs or enables the Enterprise Vault extension for users or an organizational unit.
Installs or enables the Safe Exam Browser (SEB) extension for users or an organizational unit.
Lists Context-Aware Access (CAA) access levels configured in the organization.
Lists Chrome customer profiles and their associated settings.
Lists all detectors (regex, URL list, word list) configured in the customer's organization.
Lists all Data Loss Prevention (DLP) rules configured in the customer's organization.
Lists organizational units (OUs) in the Chrome customer's directory hierarchy.
Searches Google Cloud organizations accessible to the authenticated user.
Queries Chrome Security Insights data including content transfers, URL visits, and summaries.
Queries Chrome Enterprise Security Insights data. Allows retrieving summaries and breakdowns of content transfers (uploads, downloads, prints) and URL visits. Supports filtering by event time, user, domain, or content category depending on the query action.
create_chrome_dlp_rule input schema references complex nested objects (condition, triggers array) without detailed field descriptions. 'triggers' enum-style values and 'condition' object structure undocumented, LLM must guess payload structure.
No visible output schema documentation for any tool. Responses are implicit; LLMs cannot predict what fields to extract for downstream chaining. E.g., list_dlp_rules response structure unknown, agent cannot reliably chain to get_dlp_rule.
Error handling strategy not visible in tool definitions. No recovery guidance (e.g., 'If user not found, call search_organizations first'). Tool definitions lack actionable error messages.
Parameter 'customerId' appears in 26/31 tools with description 'defaults to my_customer'. No documentation of what 'my_customer' resolves to or how override works. Implicit behavior creates LLM uncertainty about scope.
Pagination parameters (pageSize, pageToken) present in some tools (count_browser_versions, security_insights_data, get_chrome_activity_log, search_organizations) but absent in others that list collections (list_dlp_rules, list_detectors, list_customer_profiles). Inconsistent pagination strategy.
Tool risk classification (READ_ONLY, WRITE, DESTRUCTIVE) present in metadata but not exposed as MCP protocol tool annotations (readOnlyHint, destructiveHint, idempotentHint). Protocol layer cannot signal tool safety properties to client/LLM.
cep_auth tool description minimal ('Authentication bootstrapper tool for managing OAuth credentials'). No guidance on when/how to invoke, what it returns, or interaction pattern. 65+ char minimum not met for user-facing clarity.