docs(mcp): add standalone MCP user guide

[RichardHightower]

Summary

• Adds docs/MCP_USER_GUIDE.md (680 lines) as the canonical user-facing guide for agent-brain-mcp: plugin-vs-MCP decision, per-host configuration (Claude Desktop, Cursor/Windsurf, Claude Agent SDK, LangChain DeepAgents via langchain-mcp-adapters), full reference for all 7 tools / 5 resources / 6 prompts with input/output schemas pulled from agent_brain_mcp/schemas.py, three end-to-end worked examples, error mapping, cancellation, troubleshooting, and v2–v4 roadmap pointers.
• Trims the embedded MCP section in docs/USER_GUIDE.md (~90 lines) to a short summary that keeps one config snippet inline and links to the new guide.
• Refreshes agent-brain-mcp/README.md: clears the stale "Phase 0 scaffold (10.0.7)" status (v1 shipped in 10.1.0), adds pip install agent-brain-ag-mcp and a "Full guide" pointer.
• Plan-of-record saved to docs/plans/2026-05-30-mcp-user-guide.md.

Verification

Documentation-only change. Verified locally:

• task before-push — passed (lint, typecheck, 416 tests, 80% coverage).
• Tool / resource / prompt name parity — diffed against tools/__init__.py, resources/corpus.py, prompts/__init__.py. Counts match (7 / 5 / 6); names byte-exact.
• Schema parity — spot-checked search_documents and index_folder input tables against agent_brain_mcp/schemas.py. Defaults and constraints (e.g. top_k: int = 10 (1-100), alpha: float = 0.5 (0.0-1.0)) match field-for-field.
• Internal cross-links — every ./PLUGIN_GUIDE.md, ./USER_GUIDE.md, ./API_REFERENCE.md, ./CONFIGURATION.md, ./GRAPHRAG_GUIDE.md, ./plans/..., and ./roadmaps/mcp/... reference resolves to a real file; the one anchor (#multi-project-support) is verified against USER_GUIDE.md.

Test plan

• Render docs/MCP_USER_GUIDE.md in the GitHub web view; confirm tables and fenced JSON blocks display cleanly.
• Click through every cross-link from the new guide (Related docs section + inline references) and confirm each lands on the right page / section.
• Spot-check the Claude Agent SDK and langchain-mcp-adapters snippets against the current upstream READMEs — the API shapes were fetched fresh via ctx7 (/anthropics/claude-agent-sdk-python, /langchain-ai/langchain-mcp-adapters) but worth a sanity check before publication.
• Optional: run the §End-to-end worked example Feat/phase1 finalization and qa gate #1 against a live local backend (agent-brain start → MCP tools/call index_folder → tools/call get_job) to confirm the JSON-RPC shapes are byte-exact.

🤖 Generated with Claude Code