v0.1.9

Updated

• Render the CCS MCP failed-run fallback block in result-first order so the fallback search result appears before the preserved raw CCS error
• Extend verification to assert the new result-first guidance text and fallback-result-before-error section order
• Sync README, design, and TODO wording to the stronger failure-fallback visibility contract without changing the existing PostToolUseFailure -> additionalContext transport

Notes

• Successful CCS MCP runs still use PostToolUse -> updatedMCPToolOutput
• Failed CCS MCP runs still use PostToolUseFailure -> additionalContext
• This release improves visibility only; it does not change MCP ownership or claim failed-run updatedMCPToolOutput support

v0.1.8

Summary

• add CCS MCP WebSearch failure fallback through PostToolUseFailure
• keep successful CCS MCP runs on the bounded dual-output path via updatedMCPToolOutput
• sync README, design, changelog, TODO, phase, patch, install, uninstall, settings example, and verification coverage

Verification

• bash ./verify.sh --target all
• user reload + in-session testing passed before push/release

v0.1.7

Summary

• add a PostToolUse MCP companion hook so CCS MCP WebSearch can keep its original result while appending a claude-code-web-hooks companion result
• update install, uninstall, settings example, verification, and governance docs for the dual-output CCS MCP coexistence contract
• verify Claude Code, Copilot VS Code, and Copilot CLI targets still pass after the coexistence update

Test plan

• ./verify.sh --target claude-code
• ./verify.sh --target all
• ./install.sh --target claude-code --with-ccs-mcp-pass-through
• in-session mcp__ccs-websearch__WebSearch run shows both the preserved CCS result and the appended companion result

🤖 Generated with Claude Code

v0.1.6

Summary

• add explicit non-blocking coexistence support for the CCS MCP WebSearch tool mcp__ccs-websearch__WebSearch
• keep native WebSearch as the only substitution path owned by this repo
• add optional --with-ccs-mcp-pass-through install/uninstall support instead of claiming the MCP matcher by default
• add governed phase + patch artifacts and sync README/design/changelog/TODO/verification for the coexistence boundary

Verification

• ./verify.sh --target claude-code
• ./verify.sh --target all
• verified allow-only pass-through for mcp__ccs-websearch__WebSearch
• verified no substituted MCP results and no duplicate provider execution in the MCP path

Commit

• 47301ae Add CCS MCP WebSearch coexistence.

v0.1.5

Summary

• add explicit copilot-cli target support alongside claude-code, copilot-vscode, and all
• adapt the Copilot wrapper pair to handle both VS Code payloads (tool_name / tool_input) and Copilot CLI payloads (toolName / stringified toolArgs)
• switch the repo hook config under .github/hooks/claude-code-web-hooks.json to official Copilot CLI shape with version: 1 and preToolUse
• sync install, uninstall, verify, README, design, TODO, changelog, and phase docs with the shipped Copilot CLI compatibility rollout

Verification

• ./verify.sh --target copilot-vscode
• ./verify.sh --target copilot-cli
• ./verify.sh --target all

Commits

• 13a1e6c Add Copilot CLI target compatibility.
• 0a1a34d Finalize release sync after Copilot CLI rollout.

v0.1.4

Summary

• add target-aware install, uninstall, and verify support for claude-code, copilot-vscode, and all
• add Copilot-on-VS-Code compatibility wrappers for WebSearch and WebFetch
• add workspace Copilot hook example under .github/hooks/
• update README, design, changelog, TODO, and phase planning for the multiple-target rollout

Verification

• ./verify.sh --target all passes
• Claude Code path remains supported
• Copilot wrapper compatibility probe passes
• WebFetch extraction backends continue to support ordered fallback with native escape behavior

Commit

• f68665e

v0.1.3

Summary

• add selectable WebFetch extraction backends across WebSearchAPI.ai Scrape, Tavily Extract, and Exa Contents
• support one active backend per request with ordered fallback across the remaining providers
• preserve native Claude WebFetch as the final escape hatch
• add shared extraction-provider contract, policy, config helpers, and provider adapters
• refine WebFetch template-heavy detection for low-text structured portal pages
• update install/uninstall/settings/verify/docs/phase files for the new runtime model

Verification

• fixture classification remains correct for article/template-heavy/browser-shell cases
• extraction provider policy checks now cover PRIMARY selection, random no-PRIMARY selection, and ordered fallback behavior
• real-key smoke completed for Tavily Extract and Exa Contents

Notes

• this release updates the repo/runtime model and documentation for three interchangeable WebFetch extraction backends
• commit: 295a52a

v0.1.2

Overview

This release expands the search layer with Exa support, simplifies public search mode guidance, and normalizes timeout/probe environment names while keeping backward compatibility for legacy names.

Added

• Exa Search provider adapter
• Exa configuration examples in settings.example.json
• richer parallel search-result behavior that returns all successful provider results and reports partial provider failures

Updated

• public search-mode guidance now focuses on fallback and parallel
• README now explains fallback and parallel step-by-step
• provider-order and SEARCH_PRIMARY behavior are documented more clearly
• phase docs and verification flow were synchronized with the current implementation state

Env naming improvements

Preferred env names now include:

• CLAUDE_WEB_HOOKS_SEARCH_TIMEOUT
• TAVILY_SEARCH_TIMEOUT
• EXA_SEARCH_TIMEOUT
• WEBFETCH_PROBE_TIMEOUT
• WEBFETCH_PROBE_MAX_HTML_BYTES
• WEBFETCH_SCRAPER_TIMEOUT

Backward compatibility

Legacy timeout names are still accepted for now, so existing configs do not break immediately.

Current defaults

• CLAUDE_WEB_HOOKS_SEARCH_MODE=parallel
• CLAUDE_WEB_HOOKS_SEARCH_PROVIDERS=tavily,websearchapi

Notes

• native fallback remains fully permissive when custom execution cannot complete successfully
• WebFetch still uses the current WebSearchAPI scraper fallback path
• Exa is currently added as a search-layer provider

v0.1.1

Overview

This release advances Claude Code Web Hooks from a single-provider search path into a documented multi-provider search architecture.

Highlights

• Added multi-provider WebSearch architecture
• Added Tavily Search support
• Added provider policy layer with single, fallback, and parallel modes
• Set current default search behavior to:
  • CLAUDE_WEB_HOOKS_SEARCH_MODE=parallel
  • CLAUDE_WEB_HOOKS_SEARCH_PROVIDERS=tavily,websearchapi
• Preserved fully permissive native fallback behavior

Included in this release

• role-based hook entry filenames:
  • websearch-custom.cjs
  • webfetch-scraper.cjs
• shared provider modules:
  • failure policy helper
  • search provider contract
  • search provider policy
  • WebSearchAPI.ai adapter
  • Tavily Search adapter
• phase planning workspace for multi-provider implementation
• updated README, design, changelog, TODO, settings example, installer, uninstaller, and verification script

Why it matters

This release improves Claude Code behavior in custom-endpoint workflows where native/server-side search processing may not be available or may be unreliable.

Notes

• WebSearch now supports WebSearchAPI.ai and Tavily Search
• WebFetch still uses the current WebSearchAPI scraper fallback path
• Architecture remains open to future search/extract provider expansion

v0.1.0

Overview

First public release of Claude Code Web Hooks.

This project is a Claude Code-only hook layer that improves practical WebSearch and WebFetch behavior when native/server-side behavior is not available or not sufficient through custom endpoint workflows.

Included

• WebSearch substitution hook using WebSearchAPI.ai
• WebFetch auto-detect + scraper fallback hook
• fully permissive fallback policy so custom-path failures fall through to native Claude Code tools
• API key pool support:
  • inline | pool
  • JSON array file
  • newline-separated text file
• automatic next-key retry across the configured key pool
• installer, uninstaller, and verification scripts
• fixtures for repeatable classification checks

Use cases

• Improve WebSearch when a custom endpoint cannot complete the expected Claude Code search process server-side
• Improve WebFetch on template-heavy or browser-render-required pages
• Preserve native Claude Code behavior when the custom path should not trap the user

Notes

• Current implementation targets WebSearchAPI.ai only
• Architecture remains open to supporting other search APIs later if a better fit is identified