nodecode-dyscalculia

Dyscalculia statistics: cited prevalence figures, references, chart data and screening-test facts from dyscalculia.io. A Nodecode rewrite of Dyscalculia-io/dyscalculia-agent-plugin by Dyscalculia-io.

The original was configuration only: it pointed an MCP client at https://mcp.dyscalculia.io/mcp and shipped one skill. Here the same endpoint is spoken from Lisp, and its six read-only tools are six tools a model can call:

Tool	What it answers
`get_prevalence`	Published prevalence figures, one per population study or guideline: the figure or range in percent, sample size, year, stated age and the reference key. Filters: `study`, `age`, `min_pct`, `max_pct`.
`list_studies`	The 28 sources the statistics page cites, with DOI or PMID, link, kind and the topics each supports; filter by `topic`.
`get_reference`	One full reference record by key (93 keys work), with DOI or PMID and the entries built from it.
`list_chart_data`	The 7 chart and table datasets (persistence, maths anxiety, interventions, by-country, sex ratio, comorbidity) with their rows and CSV, JSON and SVG addresses.
`explain_screening_test`	The five number-skills screening games, what each measures and how they are scored; `include_items` adds the item sets.
`get_citation`	How to cite and reuse the compilation: the licence, the citeAs text, the credit link, the downloads.

And /dyscalculia for the operator, answered from the same six.

Get it

Tell the agent "add the dyscalculia cell", or run nodecode add nc://dyscalculia.

Set it up

Nothing is required. The server needs no key, no account and no sign-in, and the cell talks to it over the image's own HTTP call. Say /dyscalculia to try it.

Three optional members, in /settings or in the dyscalculia section of config.jsonc:

{ "dyscalculia": { "base_url": "https://mcp.dyscalculia.io/mcp", "timeout_seconds": 30, "max_chars": 8000 } }

Member	Default	Notes
`base_url`	`https://mcp.dyscalculia.io/mcp`	the MCP endpoint; https, or http on localhost for a server of your own
`timeout_seconds`	`30`	how long one call may take
`max_chars`	`8000`	the longest answer handed to the model; longer text is cut at that many characters and says so

A member of the wrong shape refuses the start and names itself; then (restart-cells "nodecode-dyscalculia").

Use it

Ask the question you actually have: "how common is dyscalculia at 11?", "what is the source for that figure?", "does it persist into adulthood?", "what does a dyscalculia test measure?". The model picks the tool, and the answer is the site's own text: figures with their studies, references with DOI or PMID, chart rows with their notes.

From eval, the same six calls:

(dyscalculia:get-prevalence :age 11)
(dyscalculia:get-prevalence :min-pct 3 :max-pct 6)
(dyscalculia:list-studies :topic "persistence")
(dyscalculia:get-reference "shalev2000")
(dyscalculia:list-chart-data :id "chart-1")
(dyscalculia:explain-screening-test :include-items t)
(dyscalculia:get-citation)

(help :dyscalculia) answers the guidance the model reads, taken from the original's skill; *primer* is that text, so a layer can reword it.

What the site asks of anyone quoting it is in that primer: give a range across studies rather than one number, cite the study behind a figure, and never state a figure the tools did not return. The screening test is a screening of basic number skills, not a diagnosis, and the server scores no one.