Model Context Protocol Workspace

One context.
Every agent.

TODOX is an MCP workspace for AI agents. Capture tasks, decisions, and lessons so every agent works from the same context — no matter which model or runtime it uses.

Agent Workspace
  • Scrape competitor pricing Done
  • Draft API migration plan In Progress
  • Review security audit Queued
  • Document MCP schema v2 Done
3 Agents Active Synced 2s ago
Capabilities

Built for agents,
not humans alone.

TODOX structures work so AI agents can read, write, and reason about your project's state through the Model Context Protocol.

01

Task Graph

Dependencies, priorities, and statuses form a queryable graph. Agents know what's blocked, what's ready, and what they should do next.

02

Decision Log

Every architectural choice, trade-off, and rejection is timestamped and searchable. New agents read history instead of repeating it.

03

Lesson Memory

Capture failures, fixes, and insights as structured data. Agents retrieve relevant lessons before attempting similar work.

04

Live Sync

MCP-native real-time updates mean every agent sees the same state. No stale context, no conflicting edits, no drift.

05

Multi-Runtime

Connect Claude, GPT, local models, and custom runtimes through a single protocol. TODOX is the neutral layer.

06

Local-First

Your data lives on your machine. SQLite, filesystem, or self-hosted. Cloud sync is optional, not mandatory.

Protocol

Model Context Protocol,
made concrete.

TODOX exposes tasks, decisions, and lessons as MCP resources and tools. Any agent that speaks MCP can interact with your workspace.

R

Resources

Structured documents that describe current state. Agents read resources to understand context before acting.

// GET mcp://todox/tasks { "id": "tsk_84k2", "title": "Refactor auth", "status": "in_progress", "priority": "high" }
T

Tools

Actions agents can invoke. Create tasks, log decisions, record lessons — all through standardized tool calls.

// CALL mcp://todox/log_decision { "context": "auth-flow", "choice": "JWT over sessions", "reason": "Stateless scales" }
Demo

See it in action.

A live terminal session showing an agent querying TODOX, receiving context, and updating state.

todox-cli — agent session
$ todox agent init --model claude-sonnet-4 Connected to workspace: acme-corp Loaded 14 tasks, 3 decisions, 7 lessons $ todox agent query "What should I work on?" → Priority: high | Status: ready | Blockers: none Task: "Migrate auth to JWT" (tsk_84k2) Relevant decision: "JWT over sessions" (2026-05-08) Relevant lesson: "Token expiry needs grace period" (les_12a9) $ todox task update tsk_84k2 --status in_progress Updated. 2 agents notified.
Architecture

Built on solid ground.

SQLite
+ Filesystem
MCP
Native Protocol
Local
First Design
Open
MIT License

Ship agents that
remember everything.

Get started free. No credit card. Local-first by default.