| name | ob-search |
| description | Search open-brain memory for past observations, session summaries, and learnings. Use when: "search memory", "what did I learn", "past observations", "past sessions", "what happened before", "previous context", "recall", "remember", "erinnere dich", "was habe ich gelernt", "letzte Session", "memory context".
|
| version | 0.1.0 |
| requires | ["standard:english-only","mcp:open-brain"] |
open-brain Memory Search
Efficiently search and retrieve memories from open-brain using the 3-layer workflow.
3-Layer Workflow (ALWAYS FOLLOW)
Layer 1: Search (discovery)
Use mcp__open-brain__search to find relevant memories. Returns compact index with IDs.
search(query="<your search term>", project="<project name>", limit=20)
Filters available: type, date_start, date_end, file_path, order_by
Common types: observation, session_summary, discovery, decision, bugfix, feature, refactor, change
Layer 2: Timeline (context)
Use mcp__open-brain__timeline to get context around interesting results.
timeline(anchor=<memory_id>, depth_before=5, depth_after=5)
Or by date range:
timeline(date_start="2026-03-01", date_end="2026-03-09", project="my-project")
Layer 3: Get Observations (full details)
Use mcp__open-brain__get_observations to fetch full details ONLY for filtered IDs.
get_observations(ids=[123, 456, 789])
Rules
- NEVER skip to Layer 3 — always start with search to find relevant IDs first
- Filter before fetching — use search results to pick only the IDs you need
- Token budget: ~50-100 tokens per search result, ~500-1000 tokens per full observation
- 10x savings: The 3-layer approach uses 10x fewer tokens than fetching everything
Example Workflows
"What did I learn about X?"
search(query="X", type="discovery") → get IDsget_observations(ids=[...]) → read full details
"What happened in my last session?"
search(type="session_summary", limit=3, order_by="newest") → recent summariesget_observations(ids=[...]) → full summary text
"Show me all decisions in project Y"
search(type="decision", project="Y") → decision indextimeline(anchor=<most_relevant_id>) → context around itget_observations(ids=[...]) → full decision details
"What was I working on last week?"
timeline(date_start="2026-03-02", date_end="2026-03-09") → browse by date
Other Useful Tools
mcp__open-brain__search_by_concept(query="...") — pure semantic/vector searchmcp__open-brain__get_context(project="...") — recent session summariesmcp__open-brain__stats() — database overview (counts, types, DB size)
