MenteDB MCP Server
Beta — MenteDB is under active development. APIs may change between minor versions.
The MCP (Model Context Protocol) server for MenteDB, the mind database for AI agents.
Potencialize seu agente
O comando do próprio publicador. Não é um endpoint do PlugMCP.
Host a copy with our keys and logs.
npx -y [email protected]O Claude Code não tem link de instalação para um comando local. Copie o comando abaixo e execute-o em um terminal.
claude mcp add mentedb-mcp -- npx -y mentedb-mcp@0.5.49Execute em um terminal.
{ "mcpServers": { "mentedb-mcp": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}claude_desktop_config.json
URL remota ou um comando na sua máquina? Como as duas instalações diferem
Rodar no PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
AbrirThis MCP server lets any AI agent (Claude, GPT, Copilot, or any MCP compatible client) use MenteDB as persistent memory. Fresh installs run a local embedded database by default; once you log in it switches to MenteDB Cloud, which removes local file locks and syncs across sessions and devices.
Install and configure in one command:
npx mentedb-mcp@latest setup copilot
Then authenticate:
npx mentedb-mcp@latest login
Login is optional: without it, everything runs locally (see Local mode). On a remote or SSH session the browser cannot reach the CLI callback; after you authorize, the dashboard shows a connection code to paste into the waiting terminal.
That's it. Your agent now has persistent memory that works across all your sessions and devices. Replace copilot with cursor or claude for other editors.
For Claude Code (the CLI), MenteDB integrates through lifecycle hooks rather than MCP tools:
npx mentedb-mcp@latest setup claude-code
This writes six hooks into ~/.claude/settings.json (and every other Claude Code config directory you use). The npx launcher always runs the latest version, and the hooks keep themselves current: when a new release adds a hook event, the next session start registers it automatically, so setup is a one time step.
Why hooks beat MCP for memory:
~/.mentedb/ and the turn proceeds normally. Recall in UserPromptSubmit gives up after 8 seconds (override with MENTEDB_HOOK_RECALL_BUDGET_MS)The hook backend follows your login state: cloud when authenticated (each hook is a single HTTP call), otherwise a local daemon that owns the embedded database and starts automatically on first use (mentedb-mcp daemon). The daemon keeps the embedding model loaded and flushes to disk after every stored turn.
Once logged in, the MCP server runs as a thin HTTP client — all memory operations (store, search, recall) are handled by MenteDB Cloud. This means:
If you prefer to run entirely offline without cloud:
mentedb-mcp --local
In local mode, the server uses an embedded database at ~/.mentedb/. Multiple processes can share it safely: writes are serialized with a cross-process file lock (flock) and reads are lock-free.
If you prefer building from source instead of npx:
cargo install mentedb-mcp
mentedb-mcp setup copilot
mentedb-mcp login
After upgrading, instructions auto-update on server startup. To manually review and confirm changes:
mentedb-mcp update copilot
The update command shows you the exact instructions that will be written and asks for confirmation. If you've customized the MenteDB block, it warns you and creates a .bak backup. Your own instructions outside the MenteDB block are always preserved.
npx mentedb-mcp@latest login
This opens your browser to authorize the CLI. Once authenticated, credentials are saved to ~/.mentedb/cloud.json and the MCP server connects to MenteDB Cloud on subsequent runs.
To check your connection:
npx mentedb-mcp@latest status
To revoke access:
npx mentedb-mcp@latest logout
You can also revoke sessions from the web dashboard at app.mentedb.com.
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS), ~/.config/claude/claude_desktop_config.json (Linux), or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"mentedb": {
"command": "npx",
"args": ["-y", "mentedb-mcp@latest"]
}
}
}
Add to your Cursor MCP configuration:
{
"mcpServers": {
"mentedb": {
"command": "npx",
"args": ["-y", "mentedb-mcp@latest"],
"transportType": "stdio"
}
}
}
Add to ~/.copilot/mcp-config.json:
{
"mcpServers": {
"mentedb": {
"command": "npx",
"args": ["-y", "mentedb-mcp@latest"],
"alwaysAllow": [
"process_turn", "store_memory", "search_memories", "forget_memory"
]
}
}
}
The alwaysAllow list lets memory tools run without approval prompts.
By default the server exposes only the essential tools: 5 in cloud mode (after login), 4 in local mode, where store_memories is not available.
The hosted endpoint https://api.mentedb.com/mcp (used by the Claude plugin and custom connectors) exposes a larger set: the five above plus list_memories, get_recent_activity, get_session_context, get_injection_context, get_action_rules, ingest_agent_file, get_ingest_status, record_injection_outcome, and store_memory_multimodal.
Pass an agent_id (any stable UUID per agent) to process_turn and store_memories and each agent recalls only its own memories plus shared ones (stored without an agent). Omit it and everything stays globally visible, matching single agent behavior. A coding agent and a research agent sharing one database no longer contaminate each other's context.
Hitting a monthly limit never breaks recall: reads keep working, new turns are served read only, and the injected context carries a notice so the assistant can tell the user. Upgrading unblocks instantly.
process_turn returnsIn local mode, when an LLM provider is configured, process_turn automatically triggers a background enrichment pipeline that enhances your memory graph over time:
Enrichment is fully automatic — no additional tools or configuration needed beyond setting an LLM provider. Results feed directly into future process_turn context retrieval, improving recall quality over time.
Configure an LLM provider via environment variables (the provider defaults to mock, which disables enrichment, so it must be set explicitly):
# OpenAI
export MENTEDB_LLM_PROVIDER=openai
export MENTEDB_LLM_API_KEY=sk-... # falls back to OPENAI_API_KEY
# Or Anthropic
export MENTEDB_LLM_PROVIDER=anthropic
export MENTEDB_LLM_API_KEY=sk-ant-...
# Or Ollama (local, no key needed)
export MENTEDB_LLM_PROVIDER=ollama
In cloud mode enrichment runs server side and needs no configuration.
Without an LLM provider, the MCP server works perfectly — enrichment simply doesn't run.
Set scope: 'always' when the user says "always remember this" or states a hard constraint.
Even if process_turn fails or errors on a turn, always call it again on the next turn. Never skip because of a prior failure.
--full-tools)In local mode (--local), you can expose all 32 tools with --full-tools for advanced memory operations including knowledge graph traversal, consolidation, cognitive systems, and GDPR forget.
mentedb-mcp [OPTIONS]
Options:
--local Force local mode (embedded database; safe for concurrent processes via file locking)
--data-dir <PATH> Data directory path [default: ~/.mentedb]
--embedding-dim <DIM> Dimension of the hash fallback used only if the local embedding model fails to load [default: 128]
--llm-provider <PROVIDER> LLM provider for local extraction: openai, anthropic, ollama, mock [default: mock]
--llm-api-key <KEY> API key for the LLM provider (overrides env var)
--llm-model <MODEL> Model name override for the LLM provider
--full-tools Expose all 32 tools (local mode only, default: 4 essential tools)
-h, --help Print help
The server writes logs to both stderr and a rolling file at ~/.mentedb/mentedb-mcp.log.
Cloud mode (after login): The server runs as a lightweight HTTP proxy on stdio transport. Memory operations are forwarded to MenteDB Cloud, which runs the MenteDB engine on ECS Fargate with per-user data directories on encrypted EFS, embeddings via Cohere embed-english-v3 (1024 dimensions) on AWS Bedrock, and server-side LLM extraction. No memories are stored locally; only credentials, logs, and the hooks' retry spool live under ~/.mentedb/.
Local mode (--local): The server uses the full MenteDB engine with its embedded storage engine, local Candle embeddings (all-MiniLM-L6-v2, 384 dimensions), and optional LLM extraction. This mode supports all 32 tools including knowledge graph, consolidation, and cognitive systems.
Found a bug or have a feature request? Open an issue.
Apache-2.0
MCP server for MenteDB, persistent memory for AI agents
{ "mcpServers": { "mentedb-mcp": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}Ferramentas
Ainda não há lista de ferramentas. Servidores remotos recebem uma após a primeira verificação de saúde.
Perguntas frequentes
Copie nesta página a configuração stdio para Cursor, Claude Desktop, Claude Code ou VS Code. Ela usa a instalação do publicador, não um proxy do PlugMCP.
Este registro não lista nenhuma variável de ambiente obrigatória.
Mentedb roda localmente via stdio.
Última verificação: 2026-11-12. MCP server for MenteDB, persistent memory for AI agents