AI & MCP

AI that actually knows your tasks

Todo doesn't just store — it indexes, understands and answers. Vector search, Claude-sourced answers, and a native MCP server for your agents.

Semantic search & Ask

Vector search

Tasks, descriptions, attachments and notes indexed in PostgreSQL + pgvector.

Ask (RAG)

A natural-language question, an answer written by Claude — with the exact sources.

Transparency

Every AI call is logged: model, prompt, input and output tokens.

Pluggable embeddings

Stay fully offline, or plug in a premium embedding provider. It's a single environment variable.

Local

all-MiniLM-L6-v2 (384 dim). Offline, no API key, your data never leaves the server.

Voyage

voyage-3 (1024 dim). Higher-quality embeddings via VOYAGE_API_KEY.

OpenAI

text-embedding-3-small (1536 dim) via OPENAI_API_KEY.

Model Context Protocol

Your tasks become tools for AI

Todo ships an MCP server (streamable HTTP + stdio). Claude and other LLM agents can create, search and update your tasks directly — securely, via a scoped API key.

task_createtask_listtask_gettask_updatetask_deletetask_searchtask_asktask_quick_add

Connect Claude in one command

# HTTP transport
claude mcp add --transport http mindlog-todo \
https://todo.mindlog.today/mcp \
--header "Authorization: Bearer mlt_…"
# Or over stdio
MINDLOG_API_KEY=mlt_… \
node packages/server/dist/mcp-stdio.js

API keys carry the mlt_ prefix and can be revoked at any time.

Give your AI a real task manager

Semantic search, sourced answers and agentic access via MCP — all on your own infrastructure, with your own Claude key.