Responsive MCP management UI for DeepSeek Harness Web — floating panel, JSON import, and profile-backed Host Remote management.
Critical failures across naming, descriptions, and schemas. Tool names lack action verbs (list, status, enable, disable are acceptable but many are vague: 'tools', 'builtins', 'reveal', 'previewImport'). Descriptions are present but extremely generic (10-50 chars, well below the 194-char baseline). Input schemas visible but severely underspecified: most accept generic 'object' types with minimal field documentation. No output schemas documented. Parameters lack type constraints, enums, and validation guidance. Error handling absent from visible code. Security concerns: 'reveal' tool exposes sensitive config without clear permission gates. Overall, this server reads as a thin wrapper around internal APIs with minimal LLM-optimization.
Add a new MCP server configuration
Add an MCP server configuration to a workspace
Get the catalog of built-in MCP servers
Disable an enabled MCP server
Enable a disabled MCP server
Get the MCP configuration view for a specific workspace
Import MCP server configurations from JSON
Generic object-type parameters with no field documentation. Tools like 'add', 'update', 'importJson' accept 'spec' or 'payload' as bare objects with only a description string. LLMs cannot infer required fields, types, or constraints.
Descriptions are 10 - 50 characters, far below the 194-char baseline. Examples: 'List all MCP servers with their current status and configuration' (60 chars) is acceptable, but 'Get available tools from MCP servers' (36 chars) lacks context on when to call it vs other discovery tools. No guidance on prerequisites, return structure, or error recovery.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | <=2025-11-25 | v2 |
Import MCP server configurations to a workspace from JSON
Install built-in MCP servers
Install built-in MCP servers in a workspace
List all MCP servers with their current status and configuration
List all available workspaces
Preview the result of importing MCP server configurations from JSON
Preview the result of importing MCP server configurations to a workspace from JSON
Get the current project connections view
Reconnect to a specific MCP server
Remove an MCP server configuration
Remove an MCP server configuration from a workspace
Reveal sensitive configuration details for a server
Reveal sensitive configuration details for a workspace server
Set the list of globally excluded MCP servers for a workspace
Get the current status of a specific MCP server by name
Get available tools from MCP servers
Update an existing MCP server configuration
Get update hint information
Update an MCP server configuration in a workspace
Get what's new information for the current version
No output schemas documented. Callers cannot know what fields to expect from list(), status(), tools(), or workspace queries. This forces LLMs to guess field names for downstream tool calls, increasing errors and wasting tokens on exploratory calls.
Vague tool names reduce discoverability. 'tools' (what tools? from where?), 'builtins' (builtin what?), 'reveal' (reveal what?), 'previewImport' (preview what import?). LLMs struggle to distinguish intent without action verbs and clear nouns.
'reveal' and 'revealWorkspaceServer' expose sensitive configuration without visible permission checks or audit logging. Code shows REDACTED_VALUE constant but no enforcement. Destructive tools (removeServer, removeWorkspaceServer) lack confirmation or dry-run support.