CC 2.1.121 added "alwaysLoad": true to MCP server config — flagged servers stay in the tool registry from session start. Skills no longer need to ToolSearch-probe before using these servers' tools.
"memory": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-memory"], "env": { "MEMORY_FILE": ".claude/memory/memory.json" } }
"memory": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-memory"], "env": { "MEMORY_FILE": ".claude/memory/memory.json" }, "alwaysLoad": true // CC 2.1.121+ }
alwaysLoad is set on memory, context7, and sequential-thinking — the universally-needed T2 trio. Tavily, fal, 21st-dev-magic stay deferred (feature-specific, API-key-gated).
# STEP 0: Probe MCPs at skill start ToolSearch(query="select:mcp__memory__search_nodes") Write(".claude/chain/capabilities.json", ...) if capabilities.memory: mcp__memory__search_nodes(query="...")
# memory is alwaysLoad in .mcp.json (CC 2.1.121+, #1541) — # probe below kept as fallback for older CC: ToolSearch(query="select:mcp__memory__search_nodes") Write(".claude/chain/capabilities.json", ...) if capabilities.memory: mcp__memory__search_nodes(query="...")
Skills with inline probes touched: explore, assess, fix-issue, implement, brainstorm, verify, review-pr, cover, expect. The probe stays for CC < 2.1.121 fallback (the unknown alwaysLoad key is silently ignored on older CC). Canonical pattern updated in chain-patterns/references/mcp-detection.md + rules/probe-before-use.md.
CC 2.1.121 added claude plugin prune to remove orphaned auto-installed plugin dependencies. claude plugin uninstall <name> --prune cascades.
| Where | What |
|---|---|
| /ork:doctor | New audit step: detect orphans via claude plugin list --json; suggest claude plugin prune |
| /ork:dream | New Phase 7: Plugin Housekeeping — runs prune weekly (tracked via .claude/state/last-prune.txt) |
| /ork:setup | Phase 5 cleanup tip mentions --prune cascade and the new plugin prune command |
| MCP | alwaysLoad | Tier | Why |
|---|---|---|---|
| memory | ✓ | T2 | Used by every reflective skill (assess, explore, fix-issue, etc.) |
| context7 | ✓ | T2 | Used by implement, fix-issue, cover for live docs lookup |
| sequential-thinking | ✓ | T2 | Used by brainstorm, plan-heavy skills |
| tavily | — | T3 | Web research only; API-key gated; defer until needed |
| 21st-dev-magic | — | T3 | UI generation only; API-key gated; defer |
| fal | — | T3 | Multimodal generation; API-key gated; defer |
| agentation | — | T3 | Disabled by default; opt-in via UI annotations workflow |
| ork-elicit | — | T2 | Plugin-internal; loads on-demand from plugin entry |