← Back to the wire22 Nov 2026
The AI WireDispatch No. 093
baeke-info ▸ product· filed 8 Nov 2026 · 2 min read

Microsoft Web IQ API: Standalone Search and Grounding for AI Agents Outside Azure AI Foundry

Microsoft’s Web IQ API gives AI agents a standalone REST and MCP-based web search and grounding service outside Azure AI Foundry, with no built-in language model and caller-written answers from ranked, citable results. It covers web, news, images, videos and page browsing, with per-call pricing, API-key or Entra auth, and restrictions on citations and caching. Developers can connect it to frameworks like LangChain or coding agents via an mcp.json, though full-page access, the /auto endpoint, and higher rate limits require Microsoft approval.

Machine-drafted illustration · reviewed by a humanFIG. 01

Listen to this dispatch

Narrated by an AI-generated voice.

Microsoft's Web IQ API is a standalone web search and grounding service for AI agents, usable outside Azure AI Foundry. For developers who previously used the retired Bing Search APIs, it offers a way out of framework lock-in: Foundry’s replacement tools are tied to Foundry agents, while Web IQ is a plain REST endpoint plus a Model Context Protocol (MCP) server any model or framework can call.

Web IQ has no language model. It returns ranked, citable content, and the caller’s model writes the answer from it. Microsoft says the same platform powers Copilot and ChatGPT. Authentication is via an API key or Microsoft Entra token.

The endpoints cover web, news, images, videos and single-URL browsing. Web results return passages by default, capped at 3,000 characters per result. Full pages as text, HTML or markdown go up to 500,000 characters, but only after a support request. /browse can crawl unindexed pages, returning 202 with retryAfter while it works. /auto, a beta endpoint for typed evidence, is enabled per AppID on request and returns 403 until then.

The developer’s hands-on tests from Belgium measured 160–430 ms for web, news, images and videos through a local proxy, and about 250 ms for browsing an indexed page. Microsoft’s benchmark claims a P95 latency of 164 ms, about 2.5 times faster than the next service, with the fewest tokens.

As of 16 October 2026, web, news, images and videos cost $7 per 1,000 calls; browse costs $1 per 1,000; auto is free in beta. A paid Azure subscription is required; Azure credits do not work. The default rate limit is 15 queries per second per AppID.

There are use restrictions: developers must show citations, quote at most 200 characters per source, and cannot cache or store results. Image and video files cannot themselves be used for grounding. There is no SLA, and full-page access needs a support request.

The MCP server is stateless and exposes only the tools enabled for an AppID. The developer used it with a LangChain Deep Agent in an assistant that routes requests to the right search tool, and says adding an mcp.json to VS Code Copilot CLI or Claude Code is enough to give a coding agent web search. The tests were done through Web IQ Explorer, a local FastAPI and React app available on GitHub.

Azure setup can have issues: a personal Microsoft account signed in to the Web IQ portal but failed to load, while a work account in the same tenant worked. Some subscriptions require a Microsoft contact to enable the feature. For developers who want grounding outside Foundry, Web IQ removes the framework lock-in; the remaining limits are the missing SLA and the support requests required for full pages, /auto and higher rate limits.

Read the original at baeke.info →

End of dispatch
More on the wire
99foundry-techcommunity ▸ tooling · filed 19 Nov 2026Deployment Pattern for Foundry Toolboxes with Independent Lifecycles98foundry-devblog ▸ product · filed 10 Nov 2026Microsoft adds agentic mode to Azure Content Understanding for multi-pass document analysis97agent-framework ▸ product · filed 10 Nov 2026Microsoft ships Agent Framework builder skills for coding agents96foundry-techcommunity ▸ product · filed 9 Nov 2026Microsoft Foundry Documents Production Agent Tuning Pipeline Using Agent Optimizer