Related Servers

Many tools have overlapping purposes with only subtle distinctions: at least three PDF indexing tools (index_pdf_library, index_library_pdfs, build_pdf_knowledge_db), four+ memory write paths (remember, store_episodic_memory, add_memory_entry, save_session_summary), and two parallel skill-proposal workflows (propose/approve/reject vs draft/apply) that an agent could easily confuse. Some tools do cross-reference each other helpfully, but at 213 tools the boundaries are genuinely hard to keep straight.

The dominant snake_case verb_noun pattern (create_task, get_tasks, update_task, delete_task, backup_db) is readable and mostly consistent, but there are notable exceptions: bare verbs (remember, recall), a noun-only tool (daily_note), an underscore-prefixed private helper exposed as a tool (_obsidian_vault), and stylistic drift like metis_doctor and full_scan. The scan_ vs search_ prefix is used inconsistently for the same 'fetch papers' purpose (scan_pubmed_alerts, scan_openalex, search_semantic_scholar).

213 tools is an extreme mismatch for an MCP server surface, an order of magnitude beyond even a generous cap. The find_tools/load_tool_group/list_tool_groups meta-tools are an admission that most of the catalogue is parked to save context, which suggests the server is really several servers — memory, literature, data-cleaning, course-building, admin, DHIS2 — fused into one.

For the sprawling domain it covers — projects, tasks, ideas, journal, memory layers, literature, PDF RAG, news scanning, backups, data profiling/cleaning, courses, code repository, DHIS2, audit trail — the lifecycle coverage is impressively thorough, with full CRUD on most entities and even backup/restore plus retention. Minor gaps exist (no delete for contacts or glossary terms, some redundant pathways), but nothing that would cause agent failure within the server's intended scope.