You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Production-grade Model Context Protocol server for the ShaneBrain Pi 5 stack.
42 tools across 12 groups. FastMCP + Weaviate RAG + Ollama + Planning System.
Add a journal entry, todo, reminder, or reflection
shanebrain_daily_note_search
Semantic search across DailyNote
shanebrain_daily_briefing
Generate today's briefing from recent notes via Ollama
Group 7 -- Drafts (2)
Tool
Description
shanebrain_draft_create
AI-generate a draft (email, post, letter, etc.) with vault context
shanebrain_draft_search
Search saved drafts by topic or type
Group 8 -- Security (3)
Tool
Description
shanebrain_security_log_search
Semantic search across SecurityLog
shanebrain_security_log_recent
Chronological recent security events, filterable by severity
shanebrain_privacy_audit_search
Search PrivacyAudit records
Group 9 -- Weaviate Admin (2)
Tool
Description
shanebrain_rag_list_classes
List all Weaviate collections with object counts
shanebrain_rag_delete
Delete a Weaviate object by collection + UUID
Group 10 -- Ollama (2)
Tool
Description
shanebrain_ollama_generate
Prompt the local LLM directly with optional system prompt
shanebrain_ollama_list_models
List downloaded Ollama models with sizes
Group 11 -- Planning (3)
Tool
Description
shanebrain_plan_list
List markdown files in a planning subfolder
shanebrain_plan_read
Read a planning file's full content
shanebrain_plan_write
Create or append to a planning file (path-traversal protected)
Group 12 -- System (1)
Tool
Description
shanebrain_system_health
Ping Weaviate, Ollama, and Gateway; return latency dashboard
Transport Options
Flag
Protocol
Use case
--transport streamable_http
Streamable HTTP (default)
Docker, remote clients
--transport sse
Server-Sent Events
Legacy MCP clients
--transport stdio
Standard I/O
Direct subprocess integration
The server listens on 0.0.0.0:8100 by default. The /mcp endpoint handles MCP
JSON-RPC, and /health returns service status (HTTP 200 healthy, 503 degraded).
Quality
Pydantic v2 BaseModel with Field constraints on every tool input
MCP annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) on all 42 tools
Actionable error messages with next-step suggestions
Logging to stderr only (never pollutes MCP stdout)
Async throughout, no blocking I/O
Lifespan manager validates Weaviate + Ollama connectivity at startup
Path traversal protection on filesystem tools (Planning)
shanebrain_ prefix on all tool names to prevent conflicts
Smoke Test
# Server must be running on localhost:8100
python3 test_smoke.py
# Custom URL
python3 test_smoke.py http://192.168.1.50:8100
The smoke test initializes an MCP session, calls each tool group, and reports
pass/fail counts. Slow tools (RAG chat, briefing, draft generation) are skipped.
Files
File
Purpose
shanebrain_mcp.py
Main server (all 42 tools + health endpoint + lifespan)
test_smoke.py
Smoke test suite for all tool groups
Note: The production deployment also uses weaviate_bridge.py and health.py
as local imports. Those modules live in the Pi deployment at
/mnt/shanebrain-raid/shanebrain-core/mcp-server/.