Use when the user asks: search memory, find session, what did I discuss, previous conversation, session history, memex, recall, cross-LLM search, AI conversation history.
Resources
9Install
npx skillscat add shetengteng/memex Install via the SkillsCat registry.
Memex MCP Skill
Local-first cross-LLM session memory hub. Search, retrieve, and browse AI conversation history from Claude Code, Cursor, Codex, and OpenCode — all indexed locally with FTS5.
Setup
# One-time: install and configure
cargo install --path crates/memex-cli
memex setup cursor # or: memex setup claude-code
# Ingest sessions
memex ingestMCP Tools
| Tool | Description | Required Params |
|---|---|---|
search_memory |
FTS5 search across all sessions | query (string) |
get_session |
Get full session by ID/prefix | session_id (string) |
list_recent |
List recent sessions | — |
stats |
Show index statistics | — |
search_memory
{ "query": "authentication middleware", "limit": 5, "adapter": "cursor", "project": "my-app" }Returns matching chunks with snippet, session ID, adapter, project, and timestamp.
get_session
{ "session_id": "abc123" }Returns session metadata and all messages (role + content).
list_recent
{ "limit": 10 }Returns recent sessions with ID, adapter, project, timestamp, and message count.
stats
Returns total sessions, messages, and chunks in the index.
CLI Commands
| Command | Description |
|---|---|
memex ingest [--adapter X] |
Ingest sessions from AI tool history |
memex search <query> [--limit N] [--adapter X] [--project X] |
Search indexed sessions |
memex sessions [--recent N] |
List sessions |
memex session <id> |
Show a specific session |
memex stats |
Show statistics |
memex config show |
Show configuration |
memex config set <key> <value> |
Set configuration value |
memex setup <target> |
Configure MCP for cursor/claude-code |
memex doctor |
Run diagnostics |
memex backup <path> |
Export data archive |
memex rebuild-index |
Rebuild SQLite from Markdown |
memex mcp |
Start MCP server (stdio) |
Configuration Keys
| Key | Type | Description |
|---|---|---|
adapters.claude_code |
bool | Enable Claude Code adapter |
adapters.cursor |
bool | Enable Cursor adapter |
adapters.codex |
bool | Enable Codex adapter |
adapters.opencode |
bool | Enable OpenCode adapter |
llm.ollama_enabled |
bool | Enable local LLM summarization |
llm.ollama_url |
string | Ollama server URL |
llm.ollama_model |
string | Ollama model name |
privacy.redaction_enabled |
bool | Enable PII redaction |
privacy.skip_private_sessions |
bool | Skip private sessions |
data_dir |
string | Data storage directory |
Custom Redaction Rules
Create ~/.memex/redactions.yaml:
rules:
- pattern: "INTERNAL-\\d+"
label: internal_id
- pattern: "corp\\.example\\.com"
label: internal_domainData Model
- Sessions: collected from AI tool history directories
- Messages: individual turns (user/assistant/system/tool)
- Chunks: FTS5-indexed segments split from messages
- Markdown files are the source of truth; SQLite is a reconstructible index
Trigger Phrases
Use when the user asks: search memory, find session, what did I discuss, previous conversation, session history, memex, recall, cross-LLM search, AI conversation history.