HR management tools for candidate data operations. Loads candidate data from JSON file or Azure Table Storage and maintains it. Provides tools to list, search, add, update, and remove candidates.
The HR MCP server defines 5 tools with consistent naming conventions (ListCandidates, AddCandidate, UpdateCandidate, RemoveCandidate, SearchCandidates) that start with action verbs. However, the server has significant gaps in schema completeness, parameter validation, error handling guidance, and output documentation. Tool descriptions are present but generic (10-60 characters), and while parameters have type definitions and descriptions, there is no documented output schema for any tool. Security practices are undocumented. The server follows a reasonable architectural pattern with dependency injection and async/await, but lacks production-grade error recovery and field-mapping clarity for chaining operations.
Adds a new candidate to the list
Provides the whole list of the candidates
Removes a candidate by email
Searches for candidates by name, email, skills, or current role
Updates an existing candidate by email
No documented output schemas for any tool. Return types are inferred from code (CandidateCollection, string) but not formally declared in tool metadata. LLMs cannot plan chained operations or extract required fields without knowing what SearchCandidates, ListCandidates, or AddCandidate will return.
Tool descriptions are extremely terse (10 - 60 characters). 'Provides the whole list of the candidates' (49 chars) and 'Adds a new candidate to the list' (34 chars) lack context on when to call these tools vs similar ones, what the return structure is, or any prerequisites. Descriptions should be 50 - 200 characters with actionable guidance.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 43 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 48 | - | v1 |
No error handling guidance in tool descriptions or returns. AddCandidate returns a raw string 'Candidate with email already exists' but does not guide the LLM on what to do next (e.g. call SearchCandidates to verify). RemoveCandidate returns 'Candidate with email not found' without recovery hints. Errors should be structured with actionable next steps.
UpdateCandidate accepts optional nullable parameters (firstName, lastName, currentRole, spokenLanguages, skills) but does not document the mutually-exclusive or conditional logic. The description says 'optional' but does not clarify what happens if you pass null vs omit it, or whether at least one field must be updated.
SearchCandidates accepts a single free-form searchTerm string with no constraints, enums, or validation rules documented. The description does not specify what fields are searched (name, email, skills, role), the match algorithm (contains, exact, regex), or case sensitivity. This invites hallucinated or malformed queries.
Email parameter is used as the unique identifier for UpdateCandidate and RemoveCandidate, but no validation is documented (format, length, domain). The tool does not explain whether the email is case-sensitive, whether leading/trailing whitespace is auto-trimmed, or what happens if the email is invalid. Code trims email but this is not mentioned in descriptions.
spokenLanguages and skills parameters accept comma-separated strings with no format validation documented. Tools will silently discard empty entries (via .Where(s => !string.IsNullOrEmpty(s))) but the user and LLM are not told this. The description should state 'Comma-separated, non-empty items only' and confirm that duplicates or blank entries are stripped.
No pagination or result limits documented for ListCandidates or SearchCandidates. If there are thousands of candidates, the full list will be returned, bloating the context window and degrading LLM reasoning. Tools should accept limit/offset parameters and document the default and maximum result count.
Destructive tool (RemoveCandidate) has no confirmation or dry-run step. An LLM could permanently delete a candidate with no reversibility. Code comment notes 'All modifications are temporary and reset when the server restarts' but this is a limitation, not a feature, agents expect persistence and assume deletions are final.
No permission gates or scope declarations. Tools do not declare what permissions they require (read:candidates, write:candidates, delete:candidates). This makes least-privilege configuration impossible and audit trails unclear.