Task-Anchor-MCP

<p align="center"> </p>

<p align="center"> <strong>ADHD executive function enforcement for Claude Desktop.</strong><br> An MCP server that transforms task management from a social contract<br> into a stateful boundary Claude cannot cross without explicit tool invocation. </p>

<p align="center"> <code>Python 3.11+</code> · <code>MIT License</code> · <code>96 tests</code> · <code>14 tools</code> </p>

---

ADHD developers don't lack ideas — they lack friction between ideas. A single "while we're at it" can derail an hour of focused work into a rabbit hole that felt productive but shipped nothing.

Task Anchor adds that friction mechanically. It creates a task lock that Claude must respect, runs drift detection on every message, and forces explicit validation before a task can be marked done. Ideas that surface mid-task are parked, not lost — they go into a safe queue you can review later.

The system also tracks emotional state across sessions. If you leave frustrated and come back the next morning, it knows, and it offers re-entry strategies instead of just dumping you back into the same wall.

---

You say something → drift_detect scores it → drift? → park the idea, redirect
                                             clear? → continue working
                                                      ↓
                                              scope_validate_edit → in scope? → proceed
                                                                   out of scope? → block + offer options
                                                      ↓
                                              task_complete → evidence matches exit condition? → release lock
                                                             doesn't match? → reject, keep lock

Claude is required to follow these rules (enforced via .claude/CLAUDE.md):