dyscalculia-mcp
MCP server for the sourced dyscalculia statistics on Dyscalculia.io: prevalence
studies with sample size, year, age and reference, 93 references with DOI or PMID where the site has one, the chart and table
datasets, and a plain description of the five number-skill screening games. Ask an assistant "what share of children have
dyscalculia, and which study says so?" and it answers from sourced figures. Read-only.
Tools
Every tool is read-only (readOnlyHint). Invalid input returns isError with the valid options. No tool scores a person,
takes answers or returns a screening result: explain_screening_test describes the test, nothing more.
Use
Hosted (no install): https://mcp.dyscalculia.io/mcp. Docs for each client: https://docs.dyscalculia.io/mcp
Local:
Claude Code: claude mcp add --scope user dyscalculia -- npx -y dyscalculia-mcp.
Self-host the HTTP transport: npm run build && PORT=3000 npm run start:http (POST /mcp, stateless, /health), or build the
Dockerfile (node:22-alpine, non-root, port 3000).
Honest limits
- There is no global prevalence figure. No worldwide pooled meta-analysis of dyscalculia prevalence was found; the
"worldwide" entry is a range across population studies. Studies use different criteria and ages, so figures are not directly
comparable, and one English sample gave 1 to 6 percent depending on the cut-off.
- The figures are published research estimates, not a diagnosis of any person. The server gives no advice about a learning
difficulty and does not tell anyone whether they have dyscalculia.
- The screening test is a screening of basic number skills. It has no norm sample, no percentiles and no published accuracy
study; it does not diagnose dyscalculia and does not rule it out.
- DOIs and PMIDs are exactly those the site prints. Where the site has none, none is given.
Source of the figures: https://dyscalculia.io/dyscalculia-statistics · Method: https://dyscalculia.io/methodology
Data CC BY 4.0 (see dyscalculia-data), code MIT.