MCP stdio server for the governed workflow admin panel. Exposes tools for workspace management, planning, research, verification, rules, proposals, and repository operations.
This server exhibits severe deficiencies in tool definition quality. While 46 tools are registered, critical inspection reveals: (1) NO input schemas are visible in the provided source code, the tool definitions exist but parameter schemas are not exposed in mcp_server.py or the individual tool files; (2) descriptions are uniformly generic and under 30 characters (e.g., 'Get the current state of the workspace', 'Advance the workspace to the next phase'), failing to explain WHEN to use each tool, what they return, or how they relate to similar tools; (3) no parameter descriptions are evident, tools like workspace_advance, workspace_extend_plan, and workspace_update_subphase provide no documentation of their inputs; (4) no output schemas are documented; (5) error handling is absent, no recovery guidance, retryability hints, or actionable error messages. The tool set is large (46 items) but poorly specified. Most tools score 0-20 due to missing schemas and under-20-character descriptions. Per hard scoring rules: tools without visible input schemas score 0 on schema; tools with <20-char descriptions score 0-20 on description.
Tools (46)
rule_createwriteauth27/100
Create a new rule in the workspace
rule_deletedestructiveauth28/100
Delete a rule from the workspace
rule_getread onlyauth29/100
Get a specific rule from the workspace
rule_listread onlyauth28/100
List rules in the workspace
rule_updatewriteauth29/100
Update an existing rule in the workspace
workspace_add_verification_stepwriteauth27/100
Add a verification step to a profile in the workspace
NO VISIBLE INPUT SCHEMAS: All 46 tools lack documented input parameter schemas. Source code shows tool names and descriptions in mcp_server.py and individual tool files, but no JSON Schema definitions with parameter types, constraints, or properties. Per hard scoring rule: tools without input schemas score 0 on schema dimension.
ADD JSON SCHEMA INPUT DEFINITIONS: For every tool, define a complete input schema with parameter types, descriptions, and constraints. Example for workspace_advance: {'type': 'object', 'properties': {'workspace_id': {'type': 'string', 'description': 'The workspace ID to advance'}, 'next_phase': {'type': 'string', 'enum': ['planning', 'execution', 'review', 'completion'], 'description': 'The next phase to transition to'}}, 'required': ['workspace_id', 'next_phase']}. Baseline: A+ tools document ALL parameters with types and descriptions.
EXPAND DESCRIPTIONS TO 50-200 CHARACTERS: Replace generic descriptions with actionable ones. Example: 'Get the current state of the workspace' → 'Retrieve the current workspace state including active phase, progress percentage, and pending reviews. Call this first to understand what actions are available. Returns state, phase, pending_count, and last_updated_at.'
DOCUMENT EVERY PARAMETER: For each tool parameter, add a description that explains: what it controls, its type, valid values (enums), and format constraints. Example: 'phase (string, enum: planning|execution|review|completion): The workflow phase to filter by. Use planning to see pre-work tasks, execution for active work, review for peer review stage.'
DOCUMENT OUTPUT SCHEMAS: Explicitly define what each tool returns. Example for workspace_get_state: 'Returns {state: string, phase: string, progress: number 0-100, pending_reviews: integer, last_updated_at: ISO8601 string}. Use state and phase values in subsequent workspace_advance calls.'
GENERIC DESCRIPTIONS (<20 CHARS): Descriptions are universally trivial and under 20 characters (e.g., 'Get the current state of the workspace', 'Advance the workspace to the next phase'). They provide no information about WHEN to use each tool, what parameters are required, what the tool returns, or how it differs from related tools (e.g., workspace_update_subphase vs workspace_extend_plan are indistinguishable). Per hard scoring rule: descriptions under 20 characters cannot exceed 40 points. Per pattern:tool-description baseline: A+ tool descriptions average 194 chars and explain what, when, and return value.
NO PARAMETER DESCRIPTIONS: Tools accept parameters but provide no documentation of what each parameter does, its type, valid values, or constraints. LLMs cannot infer parameter meaning from names alone (e.g., a 'status' parameter could mean HTTP status, project status, or boolean flag). This violates pattern:tool-description which requires every input parameter to have a non-empty description.
NO OUTPUT SCHEMA DOCUMENTATION: Tools return structured data but no documentation exists describing what fields are returned, their types, or which fields downstream tools need. This forces LLMs to guess at output structure and wastes tokens on trial-and-error binding. Pattern:tool requires documented return types.
DESTRUCTIVE TOOLS WITHOUT CONFIRMATION: workspace_delete_subphase, workspace_delete_research, workspace_delete_criteria, rule_delete are flagged DESTRUCTIVE but have no mention of dry-run, confirmation step, or recovery guidance. Pattern:confirmation-request requires irreversible operations to support a dry-run or confirmation flow to prevent accidental data loss.
AMBIGUOUS TOOL DIFFERENTIATION: Tools like workspace_update_subphase, workspace_extend_plan, workspace_update_criteria, workspace_update_verification_profile, and workspace_update_progress all perform update operations but have identical 2-3 word descriptions. No guidance on when to use each. LLMs will conflate these and call the wrong tool. Naming and descriptions must distinguish overlapping tools.
NO ERROR HANDLING GUIDANCE: Tools provide no documentation of failure modes, retryability, or recovery steps. Per pattern:recovery-guide, error responses must tell the LLM what to do next. A missing field or invalid ID should guide the LLM toward a lookup or alternative approach, not dead-end.
LIST TOOLS WITHOUT PAGINATION GUIDANCE: workspace_list_research, workspace_list_proposals, rule_list, workspace_get_review_issues provide no information about pagination, result limits, or total counts. Per pattern:paginated-result, tools returning lists must accept page/offset and limit parameters and return a total count or next_cursor to prevent context window exhaustion.
ADD CONFIRMATION FLOW TO DESTRUCTIVE TOOLS: Implement a dry-run parameter or separate confirm_* tool for workspace_delete_subphase, workspace_delete_research, workspace_delete_criteria, and rule_delete. Example: 'Add confirm=true parameter; if false, return {would_delete: true, affected_count: N, preview: [...]}. Require confirm=true to execute deletion.' This prevents accidental data loss.
DISAMBIGUATE UPDATE/EXTEND TOOLS: Rename and clarify overlapping tools. Example: workspace_extend_plan (add new phases) vs workspace_update_subphase (modify existing phase details) vs workspace_update_criteria (change verification criteria). Each description must state the specific change it makes.
ADD ERROR RECOVERY GUIDANCE: Document failure modes and recovery steps. Example: 'If workspace_id not found, returns {error: 'workspace_not_found', available_ids: [...], hint: 'Try workspace_list_proposals to see active workspaces.'}. Provide actionable next steps instead of bare error codes.
DOCUMENT PAGINATION FOR LIST TOOLS: workspace_list_research, workspace_list_proposals, rule_list, workspace_get_review_issues should accept limit (default 20, max 100) and offset/cursor parameters. Return total_count and has_more. Example: 'Returns {items: [...], total_count: 347, limit: 20, offset: 0, has_more: true}. Call again with offset=20 to fetch next page.'
ADD REQUIRED PARAMETER DOCUMENTATION: Every tool parameter that doesn't have a sensible default must be marked as 'required' and explained. Example for workspace_set_plan: 'phase (required): The phase name (2-50 chars, alphanumeric + underscore). No default, plan phase must be explicitly provided.'
GROUP RELATED TOOLS BY PREFIX: Ensure tools with a shared prefix (e.g., workspace_*) have consistent naming and parameter styles. This helps LLMs recognize tool families and reduces confusion when many similar tools exist.