Mem Search

1---2name: mem-search3description: Search claude-mem's persistent cross-session memory database. Use when user asks "did we already solve this?", "how did we do X last time?", or needs work from previous sessions.4---5 6# Memory Search7 8Search past work across all sessions. Simple workflow: search -> filter -> fetch.9 10## When to Use11 12Use when users ask about PREVIOUS sessions (not current conversation):13 14- "Did we already fix this?"15- "How did we solve X last time?"16- "What happened last week?"17 18## 3-Layer Workflow (ALWAYS Follow)19 20**NEVER fetch full details without filtering first. 10x token savings.**21 22### Step 1: Search - Get Index with IDs23 24Use the `search` MCP tool:25 26```27search(query="authentication", limit=20, project="my-project")28```29 30**Returns:** Table with IDs, timestamps, types, titles (~50-100 tokens/result)31 32```33| ID | Time | T | Title | Read |34|----|------|---|-------|------|35| #11131 | 3:48 PM | 🟣 | Added JWT authentication | ~75 |36| #10942 | 2:15 PM | 🔴 | Fixed auth token expiration | ~50 |37```38 39**Parameters:**40 41- `query` (string) - Search term42- `limit` (number) - Max results, default 20, max 10043- `project` (string) - Project name filter44- `type` (string, optional) - "observations", "sessions", or "prompts"45- `obs_type` (string, optional) - Comma-separated: bugfix, feature, decision, discovery, change46- `dateStart` (string, optional) - YYYY-MM-DD or epoch ms47- `dateEnd` (string, optional) - YYYY-MM-DD or epoch ms48- `offset` (number, optional) - Skip N results49- `orderBy` (string, optional) - "date_desc" (default), "date_asc", "relevance"50 51### Step 2: Timeline - Get Context Around Interesting Results52 53Use the `timeline` MCP tool:54 55```56timeline(anchor=11131, depth_before=3, depth_after=3, project="my-project")57```58 59Or find anchor automatically from query:60 61```62timeline(query="authentication", depth_before=3, depth_after=3, project="my-project")63```64 65**Returns:** `depth_before + 1 + depth_after` items in chronological order with observations, sessions, and prompts interleaved around the anchor.66 67**Parameters:**68 69- `anchor` (number, optional) - Observation ID to center around70- `query` (string, optional) - Find anchor automatically if anchor not provided71- `depth_before` (number, optional) - Items before anchor, default 5, max 2072- `depth_after` (number, optional) - Items after anchor, default 5, max 2073- `project` (string) - Project name filter74 75### Step 3: Fetch - Get Full Details ONLY for Filtered IDs76 77Review titles from Step 1 and context from Step 2. Pick relevant IDs. Discard the rest.78 79Use the `get_observations` MCP tool:80 81```82get_observations(ids=[11131, 10942])83```84 85**ALWAYS use `get_observations` for 2+ observations - single request vs N requests.**86 87**Parameters:**88 89- `ids` (array of numbers, required) - Observation IDs to fetch90- `orderBy` (string, optional) - "date_desc" (default), "date_asc"91- `limit` (number, optional) - Max observations to return92- `project` (string, optional) - Project name filter93 94**Returns:** Complete observation objects with title, subtitle, narrative, facts, concepts, files (~500-1000 tokens each)95 96## Examples97 98**Find recent bug fixes:**99 100```101search(query="bug", type="observations", obs_type="bugfix", limit=20, project="my-project")102```103 104**Find what happened last week:**105 106```107search(type="observations", dateStart="2025-11-11", limit=20, project="my-project")108```109 110**Understand context around a discovery:**111 112```113timeline(anchor=11131, depth_before=5, depth_after=5, project="my-project")114```115 116**Batch fetch details:**117 118```119get_observations(ids=[11131, 10942, 10855], orderBy="date_desc")120```121 122## Why This Workflow?123 124- **Search index:** ~50-100 tokens per result125- **Full observation:** ~500-1000 tokens each126- **Batch fetch:** 1 HTTP request vs N individual requests127- **10x token savings** by filtering before fetching128 129## Knowledge Agents130 131Want synthesized answers instead of raw records? Use `/knowledge-agent` to build a queryable corpus from your observation history. The knowledge agent reads all matching observations and answers questions conversationally.