FAQ

Direct answers. If a question is missing, open an issue.

What is this for

The basics. What NeuroDock is, what it plugs into, and what you need before you start.

What is MCP?

The Model Context Protocol. An open standard for connecting language models to external tools and data. An MCP server exposes named tools with typed inputs and outputs; an MCP-aware client (Claude Desktop, Claude Code, Cursor) calls those tools when the model asks for them. NeuroDock ships five MCP servers — none of them are model-specific.

Do I need to know Python?

No. You need Python installed (3.11+) so the substrate’s MCP servers can run, but you do not need to read or write Python to use NeuroDock. Skills are markdown, profiles are YAML, install is one command.

Will this work without Claude Desktop?

Yes. The substrate is vendor-neutral. Claude Code and Cursor are both supported out of the box. Any MCP-aware client works — the wire protocol is the boundary, not any particular vendor. See the substrate concept page for why.

Which neurotype preset should I use if I’m AuDHD?

profiles/audhd.yaml. AuDHD is a first-class enum value in the schema, not a sum-of-parts. The preset activates ADHD-tuned skills (adhd-daily-planner, hyperfocus-formatter) AND autism-leaning skills (asd-meeting-translator, audhd-context-recovery) at the same time, with chunk-size and break cadence tuned for the combined load. See the profiles reference.

Privacy & safety

What leaves your machine, what doesn’t, and how the consent model is built.

Is my data sent anywhere?

No, by default. Local-first is the default in three independent consent gates: embeddings, telemetry, and OS idle detection. Each is opt-in per scope. Choosing cloud_* for embeddings triggers a visible session-start notice every time you launch a client. The substrate has no remote endpoint to send to. See profiles.

Here is the picture. Solid lines are what runs by default; dashed lines only happen if you explicitly turn them on.

flowchart LR
  you([You]) --> claude[Claude]
  claude --> servers[NeuroDock MCP servers]
  servers --> local[(Local files<br/>~/.neurodock<br/>SQLite + YAML)]
  servers -.->|only if you opt in| cloud_embed[Cloud embeddings<br/>Voyage / OpenAI]
  servers -.->|reserved, not used yet| telem[Telemetry endpoint]

What happens if I delete `~/.neurodock/`?

You lose your profile, your cognitive graph, and your local session state. The MCP server entries in your client configs survive (they only point at uvx-installed packages), so you can re-run neurodock init and start over with a fresh profile. Nothing in ~/.neurodock/ is sent off your machine, so there is nothing to “delete from the cloud”.

Do I need a diagnosis to use the ADHD/autistic skills?

No. Self-identification is sufficient. The profile’s identity.neurotypes field is a self-ID list, never a diagnosis claim. This is a manifesto property — see manifesto §3.

Errors & bugs

Things that trip people up most often, and what to try first.

Why doesn’t `neurodock` work after `pip install`?

The CLI is on npm, not PyPI. The Python packages (neurodock-mcp-chronometric, neurodock-mcp-cognitive-graph, etc.) install via uv or pip; the CLI installs via npx or a global npm install -g @neurodock/cli. The simplest one-liner is:

npx --yes @neurodock/cli@latest install-all

That detects uv or pip and installs the six Python MCP servers, then wires your MCP clients. See the CLI reference.

Adding my own prompts / plugins

How to extend NeuroDock with your own skills, prompts, and plugin directories.

How do I add more plugins?

Plugins live in two places: in-tree at plugins/<name>/ (review by PR) and out-of-tree at ~/.neurodock/plugins/<name>/ (no project-side review). The substrate auto-discovers any directory with a valid plugin.yaml in either location. See the plugins concept page and the bundled plugins reference for the 15 plugins shipped in v0.2.1.

FAQ