Substrate
Five MCP servers and a handful of skills that externalise executive function — time, memory, and task decomposition — so you don’t have to hold them in your head.
NeuroDock
A cognitive substrate that remembers, paces, and refuses. Built with neurodivergent professionals, not for them.
NeuroDock is an AI toolkit that handles scheduling, memory, and focus boundaries so your brain doesn’t have to.
The three pillars
Section titled “The three pillars”Communication
A translation layer that reads the subtext in incoming messages and checks the tone of outgoing ones. Same MCP contract, deterministic baseline plus optional LLM refinement.
Guardrails
A guardrail layer that refuses to amplify rumination, validate hyperfocus spirals, or sycophantically agree with you. Transparent thresholds. You stay in charge.
What you can do today
Section titled “What you can do today”All three pillars are live on main and installable from npm + PyPI. See CHANGELOG.md for the per-package version index.
Substrate
Section titled “Substrate”Five MCP servers, all published to PyPI:
mcp-chronometricv0.0.2— time, session, and break management.mcp-cognitive-graphv0.0.3— persistent entity memory and recall, local-first, with a 4-rung resolution cascade (exact → alias → fuzzy → embedding).mcp-task-fractionatorv0.0.3— decompose vague goals into atomic 5–90 minute tasks (ISO 8601 duration parsing).mcp-translationv0.0.2— deterministic-baseline-plus-LLM-envelope tools for incoming subtext, outgoing tone, and meeting briefs.mcp-guardrailv0.0.3— rumination, hyperfocus, and sycophancy detectors. All three live.
Six skills compose those servers into daily-use workflows:
adhd-daily-planner— Monday-morning brief.audhd-context-recovery—/resumefrom yesterday’s state.ocd-decision-finalizer— refuse to re-validate a closed decision.hyperfocus-formatter— “Answer First” output shaping.visual-organizer— Mermaid for overwhelm states.asd-meeting-translator— transcript → four-section brief with verbatim ambiguity quotes.
Communication layer
Section titled “Communication layer”mcp-translation v0.0.2 ships a deterministic baseline plus an LLM refinement envelope, both behind one MCP contract. Bidirectional translation of corporate communication: literal-translate incoming subtext, tone-check outgoing messages, structure meeting transcripts. The browser extension (@neurodock/extension-browser v0.0.4) wires five real LLM providers — Ollama (local default), LM Studio, Anthropic, OpenAI, and OpenRouter (including openrouter/auto) — plus a deterministic mock lane for development. Non-localhost local providers (e.g. LM Studio bound to a LAN address) are gated behind a per-host chrome.permissions runtime grant.
Guardrails
Section titled “Guardrails”mcp-guardrail v0.0.3 ships all three detectors live: rumination, hyperfocus, and sycophancy (48 tests). Transparent, user-configured, overridable. Thresholds live in the user’s profile under guardrails.
Principles
Section titled “Principles”- Lower friction for users and contributors.
- Local-first by default. Cloud is opt-in per scope.
- The user is the authority. Self-ID sufficient.
- Composable over monolithic.
- Refuse where appropriate.
Read the full manifesto for what these mean in practice.
License
Section titled “License”AGPL-3.0-or-later. The substrate, the eval corpora, the documentation. Forever.