Skip to main content

​
Purpose

mem_search is the first tool you should use when looking for relevant context from past coding sessions. It searches across all projects and returns a compact index of matching observations. Use this to find relevant IDs, then call mem_get to retrieve full details. This tool provides a lightweight way to discover what’s in memory without loading full observation details.

​
Parameters

​
query
string
required
Search query — keywords or natural language
​
project
string
Filter by project name (optional — omit to search across all projects)
​
limit
number
default:"5"
Max results to return (max: 20)

​
Response Format

Returns a text summary with:
  • Total number of matches
  • Project scope (specific project or all projects)
  • List of observations with:
    • [ID:n] - Observation ID for use with mem_get or mem_timeline
    • Date of observation
    • Tool name used
    • Related files (if any)
    • Summary text
Example response: 
Found 3 memories (all projects):

[ID:142] 2026-03-01 | bash | server.ts
  Started development server on port 3000

[ID:156] 2026-03-02 | edit | api/routes.ts
  Added authentication middleware to API routes

[ID:203] 2026-03-03 | read | config/database.ts
  Reviewed database connection configuration
If no results found: 
No matching memories found.

​
Usage Example

Search across all projects: 
{
  "query": "authentication bug",
  "limit": 10
}
Search within a specific project: 
{
  "query": "database migration",
  "project": "my-app",
  "limit": 5
}
Natural language query: 
{
  "query": "when did I fix the login redirect issue"
}

​
Tips

  • Start here first: Always use mem_search before mem_get to avoid loading unnecessary data
  • Use natural language: The query understands both keywords and natural language
  • Filter by project: Use the project parameter when you know the context is project-specific
  • IDs are reusable: Save IDs from search results to use with mem_get or mem_timeline
  • Default limit is conservative: Default is 5 results to save tokens. Increase if you need broader coverage

Build docs developers (and LLMs) love

Get started for freeTalk to us