Static source inference · medium confidence · detected: Logging
Deprecated protocol patterns detected
Summary
This server exhibits significant definition quality gaps across nearly all dimensions. While 13 tools are registered with names and basic descriptions, the majority lack sufficient parameter documentation, output schema clarity, and error handling guidance. Tool names follow verb_noun convention but descriptions are often generic and fail to explain WHEN or WHY to use tools. Parameter descriptions are minimal or absent, many tools accept vague 'object' types without field-level documentation. No evidence of output schema documentation, pagination support, or recovery guidance in error cases. The security model (install/uninstall tools, call arbitrary tools via call_tool) is present but lacks explicit permission gates, audit logging, or scope declarations. Overall patterns coverage is ~25-30%, placing this solidly in the D range (poor/major issues).
Vague parameter types (object) without field-level documentation. Tools like 'install' accept 'config' as an untyped object and 'lookup_entity' accepts 'entity_type' as string with no enum. LLMs cannot infer valid values or structure.
No output schemas documented. Tools like 'list_tools', 'list_servers', 'search' return data but no CallToolResult structure is specified in descriptions. Agents cannot plan downstream calls without knowing what fields are returned.
Descriptions are generic and lack WHEN/WHY guidance. E.g., 'Search for tools and servers in the ToolPlex registry' does not explain when to use this vs 'list_tools', or what the query syntax accepts. Baseline is 194 chars; most here are 30-60 chars.
search
Recommendations
Add enums to constrained parameters: 'entity_type' → enum: ['server', 'tool', 'integration']; 'notification level' → enum: ['info', 'warning', 'error']. Replace free-form object params with typed fields or reference structured config schema in description.
Expand all descriptions to 150 - 200 chars (baseline 194). Answer: What does it do? When to use (vs similar tools)? What does it return? E.g., 'Search the ToolPlex registry for available MCP servers and tools by keyword. Use this to discover new capabilities before installation. Returns matching servers with descriptions and install IDs.'
Document output schemas in CallToolResult. For each tool, specify returned fields and types. E.g., 'Returns {servers: [{id, name, description, install_command}], total_count, next_cursor}' or link to output documentation.
Add per-parameter descriptions and constraints. E.g., 'server_id (string, required): The unique server identifier. Format: lowercase alphanumeric + hyphens (e.g. 'github-integration'). Call list_servers() to discover available IDs.' For 'config' in install, either break into typed fields (runtime, transport, command, args) or document the expected schema inline.
Add error handling and recovery guidance. E.g., install: 'Returns error if server_id is already installed, check list_servers() first, or call uninstall() to reinstall.' call_tool: 'If tool fails, check that server is running (list_servers) and tool exists (list_tools). Retryable: true for network errors, false for invalid arguments.'
Spec posture evidence
Inferred effective spec: <=2025-11-25.
Relies on Logging (deprecated) - log to stderr or use OpenTelemetry
Score history
Overall score trend
↓ 11 points across a rubric change (v1 → v2)
34/100
Scored
Grade
Overall
Spec posture
Rubric
2026-09-22
F
34
<=2025-11-25
v2
2026-03-09
F
45
1.23.0+
v1
notifyread only42/100
Send notifications to the client
save_playbookwriteauth42/100
Save a playbook for tool automation workflows
searchread onlyauth50/100
Search for tools and servers in the ToolPlex registry
No error handling guidance. Tools do not document what errors are possible, whether they are retryable, or how to recover. E.g., 'install' may fail if a server is already installed, but this is not mentioned. Recovery patterns absent.
No pagination support documented. 'list_tools' and 'list_servers' do not mention limit, offset, or cursor parameters. If servers/tools lists are large, responses blow context window without pagination.
Parameter descriptions missing or trivial. E.g., 'entity_type' in 'lookup_entity' has description 'Type of entity to lookup', does not specify valid values (server, tool, integration?). 'server_id' throughout lacks explanation of format or resolution.
Tools combine multiple concerns. E.g., 'install' both fetches server config and registers it in one step, no separation of discovery (search) from registration (install). 'call_tool' + 'initialize_toolplex' blur setup and execution.
Ambiguous tool names. 'list_tools' vs 'search', both discover tools, but distinction unclear from names alone. 'save_playbook' + 'log_playbook_usage' sound like similar operations; no clear separation of concerns.
No idempotency guarantees stated. Tools like 'install', 'save_playbook', 'call_tool' do not declare whether repeated calls with identical inputs are safe. Agents retrying on failure risk duplicate side effects (double installation, duplicate playbooks).
installsave_playbookcall_toollog_playbook_usage
Declare permission requirements per tool. Add to descriptions: install requires 'admin' or 'write:servers'; call_tool requires 'execute:tools'; search requires 'read:registry'. Enable least-privilege agent configurations.
Add pagination support to list_* and search tools. Introduce limit (1 - 100, default 20), offset, and total_count in response. Document in description: 'Returns max 20 results. Use offset for pagination. Call again with offset += limit to fetch next page.'
Separate installation discovery from registration. Keep 'search' as read-only discovery. For 'install', require explicit server_id, name, and config parameters (no auto-fetch). If auto-fetch is needed, create a separate 'resolve_server_config' tool.
Clarify playbook tools. Rename 'log_playbook_usage' → 'record_playbook_execution' to better indicate state change. Add descriptions: 'save_playbook: Persist a playbook definition with name and description. Returns playbook_id.' 'record_playbook_execution: Log that a playbook ran, for audit and metrics. Returns execution_id.'
Add idempotency declarations. Tools like 'install' should state: 'Idempotent: calling install twice with same server_id succeeds once (second call is no-op if already installed, or returns error with guidance to uninstall first).' Similarly for save_playbook.
Document authentication and secret handling. If 'install' or 'call_tool' require user credentials or API keys, explain that they are handled server-side (not exposed as parameters). State where credentials come from (environment, config file, agent context).