Demographic and Health Surveys

Name: Demographic and Health Surveys MCP Server
Availability: InStock
Author: Demographic and Health Surveys

by Demographic and Health Surveys

Healthcare14 tools

Query 40 years of nationally representative household survey data on fertility, child mortality, nutrition, HIV and family planning across 90+ countries from your AI assistant. Read-only aggregate indicators — no individual survey records — with a data model you have to understand before you trust a comparison.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on 19 August 2026.

Anthropic states this reflects the level of review a connector received, not a security audit.

Connect Demographic and Health Surveys via MCP

https://api.dhsprogram.com/mcp

Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.

Demographic and Health Surveys Tools & Capabilities (14)

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

Confidence intervals are almost never populated. The DHS Program's field descriptor states the 95% confidence interval is only presented for the maternal mortality ratio and the HIV prevalence indicators. Every other indicator returns a point estimate.
Complex sample design means you cannot compute intervals from what the API returns. The DHS Program states accurate standard errors and confidence intervals require the primary sampling unit, stratification and weight variables — which exist in the microdata, not in aggregate indicator rows. It also states that simple weighting produces intervals assuming a simple random sample, which DHS surveys are not.
Small subgroups can be unreliable. The weighted and unweighted denominators ship on every row; the unweighted count is the number of actual interviews behind the estimate, and a small one should be reported rather than hidden.
Indicators are not automatically comparable across surveys or countries. The DHS Program builds comparability with standard model questionnaires, but states that a country can add questions of interest and delete questions irrelevant to it.
Denominator basis differs. De jure usual residents and de facto overnight residents are distinct populations; The DHS Program states more than 90 percent of listed people are typically both, and warns a person could otherwise be double counted.
Multiple values can exist for one indicator. IsPreferred marks the preferred value and IsTotal marks national totals. Ignoring either can produce a subgroup-versus-national comparison by accident.
Data may not exist for every completed survey. The DHS Program says so explicitly of its survey status values, which is why availability should be checked before rows are requested.
No microdata. Recode files require a registered account, an approved research request naming the intended analysis, per-country access, and a commitment not to redistribute.
No cluster coordinates. Geometry is region polygons. Cluster GPS is a separate gated dataset, and DHS randomly displaces it — up to 2 km urban, 5 km rural, and up to 10 km for 1% of rural clusters — before any release.
Page caps bound each request. 1,000 rows unauthenticated, 5,000 with a free key, 3,000 for the indicators list, defaulting to 100. Wide breakdowns need pagination.
No published request-rate limit was found. We could not verify one either way; The DHS Program's guidance to keep result sets bounded is the operative instruction.
Citation is expected. The DHS Program asks users to cite the API in any application, website, tool or document receiving its data, and results arrive carrying survey, country and year — which a summarizing agent can silently drop.
We could not verify tool behaviour ourselves. The endpoint returned HTTP 401 to an anonymous handshake on 2026-08-19. Tool names come from Anthropic's directory snapshot and The DHS Program's documentation, and no tool is annotated in anything we could observe.
Prompts are unconfirmed. Anthropic's snapshot lists none, and the gated endpoint blocked our own check.
Documentation is hash-routed. The published docs URL resolves client-side only, so the content on this page was retrieved by reading the application's router and fetching its page fragments directly.
Aggregate data is not a substitute for the survey report. Sampling errors, design effects and the survey-specific methodology live in each survey's final report, which the API points to but does not summarize.

Frequently asked questions

Does the DHS Program MCP server expose individual survey respondents' data?

No. Every tool reads The DHS Program's public aggregate-indicator API, which returns published survey estimates rather than individual records. The DHS Program states in its directory listing that no individual-level survey records are exposed and that microdata access remains governed by its existing registration flow. Household and person-level recode files still require a registered account and an approved research request reviewed by The DHS Program.

What does list_dhs_geometry return — survey cluster coordinates or boundaries?

Administrative boundaries. The DHS Program's geometry field descriptor names survey, country, year, region ID, a level rank determining the administrative boundary, and coordinates making up the polygon of the specified region. There are no cluster points. DHS releases cluster GPS separately as datasets requiring an account, a signed terms-of-use consent and review, and it randomly displaces every coordinate before release.

Do DHS indicator values from the API come with confidence intervals?

Rarely. The data response carries lower and upper 95 percent confidence bounds, but The DHS Program's own field descriptor states the confidence interval is only presented for the maternal mortality ratio and the HIV prevalence indicators. Everything else arrives as a point estimate with weighted and unweighted denominators, so sample size is your available signal about how much weight a figure can carry.

Can you compare a DHS indicator between two countries or two survey years?

Sometimes, and not automatically. Each value belongs to one indicator measured by one survey in one country in one year, on that survey's questionnaire, denominator and recall period. The DHS Program builds comparability through standard model questionnaires but states countries may add or delete questions. Denominators differ too — de jure usual residents versus de facto overnight population. Check the indicator definition before differencing two numbers.

Is the DHS Program MCP server free, and does it need an API key?

The connector is free and needs no API key. It is USAID-funded public infrastructure implemented by ICF. The endpoint is OAuth-gated, advertising one scope and dynamic client registration, so a standard MCP client connects without pre-provisioned credentials. The underlying REST API offers a free optional key that raises the per-request page cap from 1,000 rows to 5,000 for registered partners.

Does the Anthropic directory list every DHS Program MCP tool?

Yes. Anthropic's 2026-08-16 snapshot names 14 tools, and The DHS Program's own MCP overview page documents the same 14 by name with no extras. The listing is accurate in both directions, which is unusual in this catalogue. The directory's permissions label reads Read, and every tool The DHS Program documents is a discovery or retrieval tool against a read-only API.

How should an agent avoid asking the DHS API for the wrong thing?

Follow the funnel The DHS Program documents. Start with discovery tools when a country, survey, indicator or tag ID is unknown, normalize user-supplied names and IDs with resolve_dhs_filters, check data availability before requesting rows, and keep result sets bounded. Skipping normalization is how an agent silently queries an indicator ID that does not mean what the question asked.

Sources

The DHS Program MCP overview — https://api.dhsprogram.com/#/mcp-overview.html (fetched 2026-08-19 via the underlying page fragment at api.dhsprogram.com/html/mcp-overview.html, after reading the single-page application's router in main.js; the source for the 14 documented tool names, the read-only framing, the recommended discovery-then-fetch workflow, and the client best-practice guidance) · retrieved 2026-08-19
The DHS Program Claude connection guide — https://api.dhsprogram.com/#/mcp-claude.html (fetched 2026-08-19 via the same fragment route; the source for the Claude Code add command, the custom-connector path, the bearer-token alternative, and the tool-use guidance) · retrieved 2026-08-19
The DHS Program API terms and conditions — https://api.dhsprogram.com/#/terms.cfm (fetched 2026-08-19 via the same fragment route; the citation requirement and the programmatic citation and version endpoints) · retrieved 2026-08-19
The DHS Program indicator data API reference — https://api.dhsprogram.com/#/api-data.cfm (fetched 2026-08-19; the filter vocabulary, the four breakdown levels, selectSurveys, and returnGeometry) · retrieved 2026-08-19
The DHS Program indicator data field descriptor — https://api.dhsprogram.com/rest/dhs/data/fields (fetched 2026-08-19; the definitions of CILow, CIHigh, DenominatorWeighted, DenominatorUnweighted, Precision, IsPreferred, IsTotal, CharacteristicCategory and ByVariableLabel, including the statement that the confidence interval is only presented for the maternal mortality ratio and HIV prevalence indicators) · retrieved 2026-08-19
The DHS Program geometry field descriptor — https://api.dhsprogram.com/rest/dhs/geometry/fields (fetched 2026-08-19; the eight geometry fields including LevelRank and polygon Coordinates, establishing that this endpoint returns administrative boundaries rather than cluster points) · retrieved 2026-08-19
The DHS Program geometry API reference — https://api.dhsprogram.com/#/api-geometry.cfm (fetched 2026-08-19; the documented uses of the geometry endpoint and its output formats) · retrieved 2026-08-19
The DHS Program surveys API reference — https://api.dhsprogram.com/#/api-surveys.cfm (fetched 2026-08-19; the survey ID form, the survey types, and the available/completed/ongoing status values with the note that data may not be available for every completed survey) · retrieved 2026-08-19
The DHS Program indicators API reference — https://api.dhsprogram.com/#/api-indicators.cfm (fetched 2026-08-19; thousands of indicators per survey, and the tag and survey-characteristic filters) · retrieved 2026-08-19
The DHS Program advanced queries reference — https://api.dhsprogram.com/#/api-advancedqueries.cfm (fetched 2026-08-19; the 1,000/5,000/3,000 page caps, the default page size of 100, returnFields, and the free API-key partner benefits) · retrieved 2026-08-19
The DHS Program GPS data methodology — https://dhsprogram.com/Methodology/GPS-Data.cfm (fetched 2026-08-19 with a browser user agent, after WebFetch received 403; the cluster displacement policy, the 2 km urban and 5 km rural and 10 km 1% figures, the sub-15-metre GPS accuracy, and the account and consent requirements for GPS datasets) · retrieved 2026-08-19
The DHS Program dataset access instructions — https://dhsprogram.com/data/Access-Instructions.cfm (fetched 2026-08-19 with a browser user agent; the registration and research-request flow, the 24-48 hour review, per-country access, the signed consent for GPS/HIV/biomarker data, and the redistribution restriction) · retrieved 2026-08-19
The DHS Program data collection methodology — https://dhsprogram.com/data/data-collection.cfm (fetched 2026-08-19 with a browser user agent; the standard model questionnaire approach including that countries may add or delete questions, the 5,000-30,000 household DHS sample sizes, and the SPA, AIS and MIS survey descriptions) · retrieved 2026-08-19
The DHS Program analysis FAQs — https://dhsprogram.com/data/analysis-faqs.cfm (fetched 2026-08-19 with a browser user agent; the mandatory use of weights, the distinction between simple weighting and complex sample parameters, and the three variables required for accurate standard errors) · retrieved 2026-08-19
The DHS Program Guide to DHS Statistics, Analyzing DHS Data — https://dhsprogram.com/data/Guide-to-DHS-Statistics/Analyzing_DHS_Data.htm (fetched 2026-08-19 with a browser user agent; the two-stage stratified probability design, the 25-30 households per cluster take, and the de jure versus de facto population distinction) · retrieved 2026-08-19
The DHS Program team and partners — https://dhsprogram.com/Who-We-Are/About-Us.cfm (fetched 2026-08-19 with a browser user agent; more than 400 surveys in over 90 countries since 1984, USAID funding, and ICF implementation) · retrieved 2026-08-19
Live anonymous probe — 2026-08-19: initialize to https://api.dhsprogram.com/mcp returned HTTP 401 with www-authenticate: Bearer resource_metadata="https://api.dhsprogram.com/.well-known/oauth-protected-resource/mcp", scope="mcp:read". No tool listing and no annotations were obtainable · retrieved 2026-08-19
Live RFC 9728 protected-resource descriptor — https://api.dhsprogram.com/.well-known/oauth-protected-resource/mcp (fetched 2026-08-19, HTTP 200, 171 bytes; one authorization server, header-only bearer methods, scopes_supported of mcp:read). The root form returns a byte-identical body; the path-prefix form returns 404 · retrieved 2026-08-19
Live RFC 8414 authorization-server metadata — https://api.dhsprogram.com/.well-known/oauth-authorization-server (fetched 2026-08-19, HTTP 200, 506 bytes; same-host issuer, dynamic client registration, PKCE S256, authorization code and refresh token grants, token endpoint auth method none, and the same single scope). The path-append form returns a byte-identical body · retrieved 2026-08-19
Doc discovery — api.dhsprogram.com/robots.txt (fetched 2026-08-19, HTTP 200) allows the site generally while disallowing /rest/dhs/, /mcp, /oauth/ and /.well-known/; llms.txt returns 404; dhsprogram.com/robots.txt disallows /gps_data and /hiv_data among others. No Content-Signal header was returned by either host · retrieved 2026-08-19
Anthropic connector directory — https://claude.ai/directory/9e5b81c6-d26f-43bb-9425-2d132ccdea40 (snapshot 2026-08-16; partner tier, permissions label "Read", 14 tool names, no prompt names, the health category, ICF as author, and the description stating 90+ countries, 450+ surveys, no individual-level records exposed, and citation included in results) · retrieved 2026-08-16