Skip to content

This release includes breaking changes for platform teams planning a safe upgrade.

Published 2d MCP Search & Web
✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Summary

AI summary

Updates Highlights, https://github.com/dondai1234, and https://github.com/dondai1234/master-fetch/releases/tag/v11.2.0 across a mixed release.

Full changelog

web-search-plus-mcp 3.3 release notes

web-search-plus-mcp 3.3.0 packages the portable Web Search Plus 3.3.0 engine
for MCP clients. This release intentionally realigns the standalone package
version from 1.2.0 to 3.3.0 so the package and engine generation are obvious at
a glance. It is not a breaking change to the MCP tool surface.

Highlights

  • bounded heading-aware semantic sections for web_extract spans;
  • provenance-safe cross-provider snippet_aggregate projection;
  • additive source_type and explainable fetch_priority result hints;
  • completion-order Research with configurable quality quorum;
  • explicit preempted_after_quorum and cooldown diagnostics;
  • repaired quality_report=true projection on the canonical v3 path.

Stable MCP surface

The server still exposes exactly two tools:

  • web_search
  • web_extract

Their names, arguments, and legacy result fields remain compatible. New evidence
and diagnostics are additive. The stdio/subprocess boundary remains unchanged.

Research behavior

Research providers are harvested as they complete, but public result order stays
deterministic. Once enough unique results, contributing providers, and domains
satisfy the configured quorum, unfinished providers can be cancelled and are
reported as preempted_after_quorum rather than silently disappearing.

Explicit Research-provider lists no longer inherit the automatic-routing
allowlist. Disabled, unconfigured, and cooldown providers are still excluded and
remain visible in diagnostics.

MCP diagnostics

When quality_report=true is requested, the MCP projection now derives its
report from validated canonical results, routing receipts, provider attempts,
source diversity, and cache status. It does not issue a second provider request.

Hound / Master-Fetch attribution

The heading-aware interaction and completion-order Research ideas are inspired
by Hound/Master-Fetch v11.2.0,
an independent MIT-licensed project created and maintained by
Bishesh Bhandari (dondai1234).

The implementation is independently reworked for Web Search Plus MCP's own
provider, provenance, budget, cache, and receipt contracts. Hound is neither
bundled nor claimed as Robby's code. The optional Hound provider connects to a
separately installed loopback sidecar.

Upgrade

pip install --upgrade "web-search-plus-mcp==3.3.0"

Or pin the version in an MCP host configuration:

{
  "command": "uvx",
  "args": ["web-search-plus-mcp==3.3.0"]
}

No configuration migration is required from 1.2.0. Hosts that pin the package
must update the pin because of the intentional version-line realignment.

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Track robbyczgw-cla/web-search-plus-mcp

Get notified when new releases ship.

Sign up free

About robbyczgw-cla/web-search-plus-mcp

Multi-provider web search with intelligent auto-routing (Serper, Tavily, Exa)

All releases →

Related context

Earlier breaking changes

  • v1.0.0 Native Perplexity and Kilo Perplexity answer endpoints removed from public provider schemas.
  • v1.0.0 Public search surface reduced to 12 verified source-result providers; extraction limited to 8 providers.
  • v0.5.1 kilo_perplexity renamed to kilo-perplexity without aliasing
  • v0.5.1 Perplexity provider separated from kilo-perplexity with distinct API keys

Beta — feedback welcome: [email protected]