Reliefweb MCP server

@pipeworx/reliefweb

UN OCHA's humanitarian information service — situation reports, needs assessments, appeals and analyses from ~4,000 humanitarian organisations, plus the curated disaster register with GLIDE numbers.

Part of Pipeworx — an MCP gateway connecting AI agents to 1679+ live data sources.

Tools

• reliefwebreports(query?, country?, disastertype?, source?, theme?, format?, fromdate?, todate?, include_body?, limit?, offset?)
• reliefwebdisasters(query?, country?, status?, disastertype?, include_description?, limit?, offset?)
• reliefwebreport(reportid, maxbodychars?) — one report in full.

Auth

BYO only, and the credential is not shaped like a key.

The API is free and charges nothing. What it requires, since 1 November 2025, is a pre-approved appname query parameter on every call. An arbitrary string is refused with 403 AccessDeniedHttpException. Verified live 2026-09-17: appname=pipeworx, appname=reliefweb, appname=test all 403 on v2; v1 additionally answers 410 (decommissioned).

Request one at . ReliefWeb reviews the request and emails the approved string, which must combine your organisation name, your purpose and random characters. Pass it as _apiKey.

Pipeworx holds no approved appname. When one arrives it is PLATFORMRELIEFWEBAPPNAME; add platformKeyEnv to the pack manifest in the same change that lands the value, never before — declaring it first routes humanitarian questions to a tool that can only refuse.

ReliefWeb rate-limits by appname, so a shared one is a shared bucket.

Data sources

• POST https://api.reliefweb.int/v2/reports?appname=…
• POST https://api.reliefweb.int/v2/disasters?appname=…
• Field tables:

Things worth knowing:

against api.reliefweb.int/v1 is dead.

• v1 is gone. It answers 410 telling you to use v2. Anything written

answers 202 with a zero-byte body from server egress, browser UA or not.

• The public site's RSS feeds are not a workaround: reliefweb.int/updates/rss.xml

a bare array of filters is rejected.

• Filters take one {field, value} or an {operator, conditions[]} group —

ISO timestamps, not bare dates.

• Date filters go on date.original (publication) as {from, to} with full

back as arrays of objects; the pack flattens them to names.

• Taxonomy fields (source, country, theme, format, disaster_type) come

reliefweb_report is the one that returns full text.

• Report bodies are long. reliefweb_reports omits them unless you ask;

Verified

2026-09-17 — refusal paths only, by design. Both the no-key refusal and the upstream 403-on-unapproved-appname path were exercised live and both say "requires an API key". Query paths cannot be verified until an appname is approved.

Quick Start

Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):

{
  "mcpServers": {
    "reliefweb": {
      "url": "https://gateway.pipeworx.io/reliefweb/mcp"
    }
  }
}

What this endpoint actually serves

tools/list at https://gateway.pipeworx.io/reliefweb/mcp returns the tools in the table above plus the shared Pipeworx meta-tools — askpipeworx, discovertools, search_within, remember/recall and the rest of the gateway-wide set. So the tool count you see is larger than this table: a single-pack endpoint currently lists roughly 30 shared tools alongside the pack's own. The connection's initialize response states its exact scope, and is the authoritative answer for a given day.

Shortened. The full README is on GitHub.