An MCP server for Netwrix Access Analyzer that provides tools to query Active Directory inventory, filesystem access permissions, and sensitive data discovery from a Netwrix NAA database
The server provides 11 tools with explicit registration and visible schemas, but significant quality issues limit production readiness. Tool naming uses PowerShell-style PascalCase (e.g., 'Connect-Database', 'Get-ADEffectiveMembership') rather than the verb_noun convention LLMs typically expect. Descriptions are present but often generic or under-specified. Parameter descriptions exist but lack explicit constraints, formats, and validation guidance. Output schema documentation is missing entirely, LLMs cannot plan downstream calls because the response structure is undocumented. Critical security concern: Connect-Database accepts plaintext username and password as parameters, violating secret-injection patterns. Error handling is present in code but not documented in tool descriptions. The server operates over STDIO only, which hard-caps the score at 50 regardless of definition quality.
Connects to a specified MSSQL database server, overriding environment settings.
Discovers where sensitive data exists based on DLP matches. Returns a list of shares, criteria, and match counts.
Discovers the effective membership of groups in Active Directory, with optional filters. Returns a list of groups, members, and their nesting levels matching the criteria. Filters use SQL LIKE logic; include wildcards (%) where needed (e.g., '%Admin%', 'Domain\User%').
Retrieves a list of exceptions defined in the AD inventory, with optional filters. Shows exception name, description, associated principal (NT Account), and object class. Filters use SQL LIKE logic; include wildcards (%) where needed (e.g., '%Admin%', 'user%'). Data is sourced from the SA_ADInventory_ExceptionsView.
Gets Active Directory object permissions with optional filters for access entries, principals, and target objects. Filters use SQL LIKE logic.
Output schemas are completely undocumented across all 11 tools. LLMs cannot plan downstream calls or structure multi-step operations without knowing the response fields and types.
Connect-Database accepts plaintext username and password as tool parameters, storing them in request traces and logs. This violates secret-injection patterns and creates a security liability.
Tool names use PowerShell PascalCase (e.g., 'Connect-Database', 'Get-ADEffectiveMembership') rather than verb_noun LLM convention (e.g., 'connect_database', 'get_ad_effective_membership'). LLMs rely on standard naming to infer intent; non-standard names increase selection errors.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 50 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 41 | - | v1 |
Discovers open shares (accessible to broad groups like 'Everyone' or 'Domain Users'). Returns a list of shares and the count of folders directly within them marked as exceptions.
Finds filesystem resources where a specific trustee (Domain\Name) has access. Filters by levels down from the share root (0 = share level only).
For a given trustee (Domain\Name) and network resource path (UNC), finds the source of their access.
Shows the current database connection status.
Provides the schema definition (columns, types, keys) for a given table.
Alias for Show-TableSchema. Provides the schema for a given table.
Duplicate tools: 'Show-TableSchema' and 'get_table_schema' provide identical functionality. This forces LLMs to waste reasoning cycles deciding between them and increases the chance of misselection. One canonical name should be used.
Parameter descriptions include example values (e.g., 'e.g., '%Admin%', 'Domain\\User%'') which LLMs tend to reuse literally rather than adapting to context. Examples should be replaced with formal constraints (enums, regex patterns, format declarations).
Tool descriptions lack error recovery guidance. When a tool can fail (e.g., database not connected, resource not found, invalid filter), the description should explain how the LLM should respond: retry, ask user, or try an alternative tool.
Enum parameters are declared as free-form strings instead of JSON Schema enums. Example: 'is_inherited_filter' in Get-ADPermissions should be {type: 'string', enum: ['0', '1']} instead of a bare string with guidance in the description.
Parameter formats and bounds are missing or incomplete. Example: 'levels_down' in Get-TrusteeAccess lacks a minimum and maximum (e.g., minimum: 0, maximum: 10). Unbounded numbers allow LLMs to pass absurd values that break APIs or cause timeouts.