Settings UI + HTTP API + model-facing mcp_manager_* tools for managing MCP servers and skills in DeepSeek Harness
This server exposes 11 tools for MCP configuration management. Tool naming follows verb_noun convention (mcp_manager_*), which is good. However, descriptions are minimal (10-30 chars), parameter descriptions are sparse or missing, and input schemas lack proper type definitions and constraints. Most parameters are strings without enum constraints or format guidance. No output schemas are documented. Error handling is not visible in the provided code. The server manages critical infrastructure (MCP server configs, skills) but lacks the rigor expected for production use.
Add a new MCP server configuration
Edit an existing MCP server configuration
Export MCP server configurations
Import MCP server configurations
List all managed MCP server configurations
Remove an MCP server configuration
Restart an MCP server
Enable or disable an MCP server configuration
Parameter descriptions are missing or trivial. 'headers' and 'env' are described as 'HTTP headers as key=value lines' and 'Environment variables as key=value lines', no guidance on format, escaping, or validation. LLMs cannot infer correct syntax.
String parameters lack enum constraints or format specifications. 'transport' accepts 'streamable-http' or 'stdio' but is defined as a free-form string. 'level' has an enum but others do not. LLMs will hallucinate invalid values.
No output schemas documented. Tools like mcp_manager_mcpm_list and mcp_manager_mcpm_export return data structures, but the response format is not specified. LLMs cannot plan downstream calls or extract required fields.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2026-07-28+ | v2 |
Get the version of the dsh-mcp-manager plugin
List all available skills
Enable or disable a skill
Destructive operations (mcp_manager_mcpm_remove) lack confirmation or dry-run support. An agent could delete critical MCP server configs without a safety gate.
Error handling and recovery guidance are not visible in tool definitions. No indication of what errors are retryable, what the LLM should do on failure, or how to disambiguate partial failures.