MCP server for interacting with Red Hat Subscription Management APIs
This server exhibits critical gaps in definition quality across all dimensions. While tools are registered with FastMCP and basic descriptions exist, the implementation lacks proper parameter validation, output schema documentation, comprehensive error handling, and LLM-optimized descriptions. The naming convention includes prefixes (account_management_, subscription_management_) that add redundancy rather than clarity. Critically, tools like account_management_create_user embed hardcoded data (address fields, phone number, default permissions) that should never be baked into tool definitions. Parameter descriptions are minimal (often just 'The X' format), and there is no evidence of output schema documentation. Error handling is non-existent, the code simply calls response.raise_for_status() and returns raw API responses without any structured error guidance for LLMs.
Create User under the account from Red Hat Subscription Management - Account management
Get User Details under the account from Red Hat Subscription Management - Account management
Get Roles Associated under the account from Red Hat Subscription Management - Account management
Get User Status under the account from Red Hat Subscription Management - Account management
Assign Organization Administrator Role To User under the account from Red Hat Subscription Management - Account management
Invite Users under the account from Red Hat Subscription Management - Account management
Tool naming includes redundant category prefixes (account_management_, subscription_management_) that obscure the action verb. Names like 'account_management_list_accounts' are 37+ characters and bury the verb. Recommendation: use verb-first naming like 'list_accounts', 'create_user', 'get_user_roles'.
Parameter descriptions are minimal and non-actionable. Examples: 'The account ID' (12 chars), 'The user ID' (11 chars), 'List of user email addresses to invite' (38 chars). Descriptions under 50 chars typically lack context for LLM decision-making. Baseline: 72 chars. All params need descriptions answering: what format? What values are valid? What is this used for?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 41 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 34 | - | v1 |
List Account Information from Red Hat Subscription Management - Account management
List all the users under the account from Red Hat Subscription Management - Account management
Get Current User's Personal Information from Red Hat Subscription Management - Account management
List all systems for a user from Red Hat Subscription Management - Subscription Management
Tool descriptions are generic and under 100 characters for most tools. Example: 'List all the users under the account from Red Hat Subscription Management - Account management' (95 chars). Baseline: 194 chars. Descriptions do not answer: When should the LLM call this vs. other similar tools? What structure does it return? What are prerequisites?
No output schema documentation. Tools return raw API responses via 'return response' without documenting what fields the LLM should expect. This forces LLMs to explore unstructured data and risks hallucination. Every tool needs: 'Returns: {field1: type1, field2: type2, ...}'.
Hardcoded business logic in account_management_create_user: address (city='RALEIGH', state='NC', zipCode='27601'), phone='18887334281', default permissions=['portal_manage_cases', ...]. This data should not be in the tool definition, it should be parameterized or pulled from configuration. This forces the tool to work only for Red Hat's US operations and prevents reuse.
No error handling or recovery guidance. Code calls response.raise_for_status() and returns raw API JSON. If a user is not found, the LLM gets an HTTP 404 with no guidance. Errors must say: 'User not found. Try list_users() with the account_id first.' Pattern: recovery-guide.
No pagination support. List tools (list_accounts, list_users, list_systems) do not accept limit or offset parameters. If an account has hundreds of users, the entire list is returned unfiltered, risking context window exhaustion. Baseline: paginated results accept limit and offset and return total_count or next_cursor.
Offline token credential handling is transport-dependent and fragile. For STDIO, it reads from environment; for other transports, it reads from request headers. This pattern does not scale to SSE or HTTP and mixes concerns, credential injection should be uniform and server-side. See pattern:secret-injection.
account_management_create_user has a parameter 'username' that is optional and defaults to email.split('@')[0]. The description says 'optional, defaults to email prefix' but does not explain: What format is valid? Are usernames case-sensitive? Unique? What errors occur if a duplicate username is provided? LLMs cannot reason about these constraints without explicit documentation.
account_management_invite_user has a parameter 'user_emails' (array). The code wraps it in 'emails': [user_emails] (double-nesting), which is likely a bug. There is no validation that emails are valid. There is no description of what happens if an invite fails for one email out of ten, does the entire call fail, or does it return partial success?
No rate limiting or timeouts declared. The make_request function has a 30s timeout, but no retry logic. If the RHSM API is slow or flaky, agents will hang or fail with no recovery path. Timeouts should be explicit in tool descriptions, and retryable vs. fatal errors should be distinguished.