openchrome-mcp
Open-source browser automation MCP server. Control your real Chrome from any AI agent.
Description
Open-source browser automation MCP server. Control your real Chrome from any AI agent.
Release History
| Version | Changes | Urgency | Date |
|---|---|---|---|
| 1.9.7 | Imported from npm (1.9.7) | Low | 4/21/2026 |
| v1.9.7 | ## v1.9.7 — Headless Cookie Persistence Fix This release addresses a critical bug where headless mode failed to persist cookies and login state across restarts, causing repeated authentication prompts (e.g., Google login) that blocked automated workflows. ### Bug Fixes #### 🔐 Headless Cookie/Session Persistence ([#606](https://github.com/shaun0927/openchrome/issues/606)) **Root Cause A — Ephemeral temp profile for headless-shell** ([#609](https://github.com/shaun0927/openchrome/pull/609)) - | Medium | 4/9/2026 |
| v1.9.6 | ## v1.9.6 — Headed Mode Tool Access Fix & Test Coverage ### Bug Fixes #### `headed: true` + `profileDirectory` tabs now fully accessible via all tools (#562, #568) Previously, when navigating with `headed: true` and `profileDirectory` together, the returned tab was inaccessible via `read_page`, `interact`, and other tools — despite the navigation itself succeeding. **Root cause**: `getOrCreateWorker()` received both `port` (headed Chrome on port 9322) and `profileDirectory`, which triggered | Medium | 4/3/2026 |
| v1.9.5 | ## Highlights This release adds first-class [Codex CLI](https://github.com/openai/codex) support to openchrome. Codex CLI's stdio launcher interprets `npx` flags differently from Claude Code, causing the generic MCP setup to fail silently. This release provides a Codex-safe preset, a dedicated setup flow, and updated documentation. ## New Features ### Codex CLI MCP setup preset ([#565](https://github.com/shaun0927/openchrome/issues/565)) **Problem:** Codex CLI failed with the documented gene | Medium | 4/3/2026 |
| v1.9.4 | ## Highlights This release includes four desktop application fixes that improve sidecar process management, port configuration, Cloudflare Tunnel integration, and the build pipeline. These changes complement the v1.9.3 session sharing and headed mode improvements with desktop app stability. ## Bug Fixes ### Sidecar: Crash state tracking ([#555](https://github.com/shaun0927/openchrome/issues/555)) **Problem:** When the openchrome sidecar process terminated unexpectedly (crash, OOM, kill), the | Medium | 4/2/2026 |
| v1.9.3 | ## Highlights This release fixes a cluster of interrelated bugs that made it nearly impossible to use authenticated sessions (e.g., AWS Console, GitHub) in openchrome. Users can now share cookies from their desktop Chrome profile and use headed mode with full session access. ## Bug Fixes ### ChromePool: Cookie sync for profile instances ([#559](https://github.com/shaun0927/openchrome/issues/559)) **Problem:** When using `profileDirectory: "Default"` in MCP tools, Chrome launched with an empt | Medium | 4/2/2026 |
| desktop-v0.1.0 | ## What's Changed - fix(ci): allow partial platform failures for desktop release (931d5c0) - fix(ci): set APPIMAGE_EXTRACT_AND_RUN for Linux linuxdeploy (4237165) - fix(ci): remove tagName from tauri-action to prevent release conflict (7a675a1) - fix(ci): simplify to single unsigned Tauri build step (624ff28) - fix(ci): resolve 3 desktop build failures (2d0664b) - fix(desktop): convert icon PNGs from RGB to RGBA for Tauri build (aed4886) - fix(ci): install desktop npm dependencies before Tauri b | Medium | 4/2/2026 |
| v1.9.2 | ## What's Changed ### Bug Fix #### Chrome Profile Selection Not Forwarded to Sidecar (#550, PR #553) The desktop app's Chrome profile selector dropdown (added in v1.9.1) correctly saved the user's selection to settings, but **never passed it to the sidecar process**. The server always launched with Chrome's default profile regardless of what the user selected. **Root cause:** A wiring gap in the Tauri IPC layer — the frontend sent `profileDirectory` via `invoke("start_server")`, but the Rust | Medium | 4/1/2026 |
| v1.9.1 | ## What's New in v1.9.1 This release adds the complete desktop application infrastructure (Tauri v2), security hardening for HTTP transport, and a bugfix for headed Chrome mode. **20 PRs merged** across MCP server core, desktop backend, and desktop frontend. --- ### 🖥️ Desktop App Foundation The OpenChrome desktop app (Tauri v2) is now fully scaffolded with all backend services in place: - **Live Session Dashboard** — Two-pane UI with real-time screenshot stream, tool call feed, session si | Medium | 4/1/2026 |
| v1.9.0 | ## OpenChrome v1.9.0 A major reliability and compatibility release with **14 merged PRs**, spanning three-tier stealth/headed fallback improvements, cross-environment IDE support, and significant internal cleanup. --- ### Highlights - **Headed Chrome session integration** — Tier 3 (headed Chrome) fallback pages are now fully integrated with the session manager, enabling `read_page`, `interact`, `screenshot`, and all other tools to operate on headed fallback tabs (#485) - **Cross-environment | Medium | 3/31/2026 |
| v1.8.15 | # OpenChrome v1.8.15 ## Highlights This release eliminates **cumulative timeout stacking** — the root cause of 120s+ tool hangs when Chrome is slow — and adds **automatic CDN/WAF block bypass** with a 3-tier fallback system. --- ## Budget-Aware Timeout System (#454, #460) ### Problem When Chrome is slow but alive, sequential CDP calls inside a single tool handler each consumed up to 15s (the per-call CDP timeout). A tool like `fill_form` with 10 sequential calls could take **150s** before | Medium | 3/28/2026 |
| v1.8.14 | ## Highlights This release completes the cumulative timeout stacking fix from #440 and hardens stealth mode fingerprint defense. Together, these changes ensure that **no tool execution can hang indefinitely** — even when Chrome is slow but alive — and that **stealth mode leaves a smaller fingerprint** for bot-detection systems. --- ## Budget-Aware Deadline Enforcement (#454) ### Problem PR #449 (v1.8.13) introduced `ToolContext` with `startTime`/`deadlineMs` and utility functions `getRemain | Medium | 3/27/2026 |
| v1.8.12 | ## Highlights This release introduces **Stealth Mode v2**, a comprehensive anti-detection system that enables OpenChrome to bypass enterprise-grade bot protection systems such as Radware/ShieldSquare (Coupang), PerimeterX (Reddit), and Akamai. All enhancements activate **only when `stealth: true`** — normal operation is completely unchanged. ### The Problem Stealth mode v1 handled CDP-free tab creation and basic `navigator.webdriver` masking, sufficient for Cloudflare Turnstile. However, ente | Medium | 3/27/2026 |
| v1.8.11 | ## What's New in v1.8.11 This release addresses systemic hang issues and Unicode query matching failures across all browser automation tools. Previously, individual tool fixes left gaps that caused the same classes of failures to recur in different tools. This release takes an architectural approach — centralizing normalization, adding fast-fail guards, and integrating circuit breakers at the shared utility level. ### Hang Prevention #### Stale Page Validity Guard (#441) When the browser disc | Medium | 3/26/2026 |
| v1.8.10 | # v1.8.10 — Reliability Hardening & Error Recovery This release significantly hardens openchrome's production reliability through code review fixes, improved error recovery information, and connection-level resilience improvements. It also adds the new `page_screenshot` tool for saving screenshots to files. ## Highlights - **11 PRs merged** addressing bugs found during the Reliability Guarantee Initiative (#403) code review - **Connection-level resilience**: Tool calls during Chrome reconnect | Medium | 3/25/2026 |
| v1.8.9 | ## v1.8.9 — Reliability & Resilience Hardening This release delivers five significant reliability improvements to OpenChrome's self-healing infrastructure, hardening the server against Chrome crashes, network disruptions, and long-running operations. Every feature was reviewed by three parallel specialist agents (CDP expert, silent-failure hunter, platform reviewer) and includes post-review fixes. --- ### New Features #### Browser State Snapshot & Cookie Restore (#420) The most impactful fea | Medium | 3/25/2026 |
| v1.8.8 | ## v1.8.8 — Reliability Hardening & E2E Certification This release focuses on **connection reliability**, **CLI correctness**, and **comprehensive E2E test coverage** as part of the ongoing Reliability Guarantee Initiative. --- ### Bug Fixes #### Force-close WebSocket on Disconnect (`src/cdp/client.ts`) When `handleDisconnect` fires, the CDP client now explicitly calls `browser.disconnect()` to immediately close the underlying WebSocket transport. Previously, in-flight CDP calls could hang i | Medium | 3/24/2026 |
| v1.8.7 | ## v1.8.7 Release Notes ### Highlights This release focuses on **production reliability**: eliminating zombie Chrome processes, hardening the shutdown sequence, adding Streamable HTTP transport for daemon deployments, and introducing operational tooling (Prometheus metrics, disk monitoring, rate limiting, deployment configs). --- ### 🔧 Bug Fixes #### Zombie Chrome Process Prevention (#391, #393, #394, #396) The most critical fix in this release. Previously, Chrome processes spawned with ` | Medium | 3/24/2026 |
| v1.8.6 | ## v1.8.6 — E2E Test Harness Reliability & Review Fixes ### Summary This release resolves critical E2E test harness bugs discovered during verification of the Self-Healing Browser Automation architecture (#347). All 8 E2E test scenarios and 2116 unit tests now pass reliably across all platforms (Ubuntu, macOS, Windows × Node 18/20/22). --- ### 🐛 Bug Fixes #### E2E Test Harness (`tests/e2e/`) - **Fix `--auto-launch` flag propagation** — Chrome auto-launch is now hardcoded in `MCPClient.sta | Medium | 3/24/2026 |
| v1.8.5 | ## v1.8.5 — Self-Healing Architecture & AI Agent Continuity ### Highlights This release delivers the complete **self-healing infrastructure** for long-running Chrome sessions (#347) and extends **multi-profile support** with validation and comprehensive E2E testing (#383, #384). Nine PRs ship new MCP tools, Chrome stability improvements, and a 24-hour endurance test framework. --- ### New Features #### Connection Health Monitoring Tool (#378) New `oc_connection_health` MCP tool exposes rea | Medium | 3/23/2026 |
| v1.8.4 | ## v1.8.4 — Runtime Multi-Profile Support & CDP Fingerprint Evasion ### Highlights Two major features land in this release: **runtime multi-profile Chrome support** (use different Chrome profiles simultaneously from MCP tools) and **CDP-level anti-detection evasion** via rebrowser-puppeteer-core. --- ### New Features #### Runtime Multi-Profile Support (#375, #369) Run multiple Chrome profiles simultaneously — each on its own Chrome instance and debug port. Pass `profileDirectory` to `navig | Medium | 3/23/2026 |
| v1.8.3 | ## v1.8.3 — Stability, IDE Compatibility & Navigation Resilience ### Highlights This release focuses on **reliability under adverse conditions** and **broader IDE compatibility**. Four PRs landed since v1.8.2, addressing navigation hangs, Chrome crash recovery, and a tool visibility issue affecting third-party MCP clients. --- ### Bug Fixes #### Progressive Tool Disclosure for Unsupported IDEs (#373) OpenChrome uses progressive tool disclosure (starting with 24 core tools, expanding on dem | Medium | 3/23/2026 |
| v1.8.2 | ## v1.8.2 — E2E Test Suite Activation & Stability Improvements ### Highlights This release activates the full E2E test suite in CI, resolves Chrome kill-recovery stability issues, fixes navigation hangs, and adds comprehensive marathon testing infrastructure. --- ### E2E Test Suite (NEW) The E2E test suite with 8 scenarios is now fully operational. All scenarios were implemented in v1.8.1 but **never ran in CI** due to a missing `test:e2e` npm script. | Scenario | What It Tests | Local Res | Low | 3/22/2026 |
| v1.8.1 | ## v1.8.1 — E2E Test Harness & Stability Fixes ### Highlights This release adds a comprehensive E2E marathon test harness for validating long-running session stability with real Chrome instances, and fixes 3 stability bugs discovered during E2E testing. --- ### New Features #### E2E Marathon Test Harness (#357) A complete Jest-based E2E test infrastructure for validating openchrome's long-running session reliability: - **8 E2E scenarios** covering marathon operation, kill recovery, GC res | Low | 3/22/2026 |
| v1.8.0 | ## Highlights **OpenChrome 1.8.0** is the largest architectural release since v1.7.0. It introduces a **5-layer self-healing architecture** for 24-hour unattended browser sessions, **AI agent continuity tools** that survive Claude Code context compaction, and **enhanced stealth mode** for improved anti-bot bypass. ### Self-Healing Architecture (#347, #354) OpenChrome now automatically detects and recovers from Chrome crashes, tab freezes, event loop stalls, and process failures — without manu | Low | 3/21/2026 |
| v1.7.19 | ## v1.7.19 — Ralph Engine: Never-Give-Up Interaction Resilience This release introduces the **Ralph Engine** — a comprehensive interaction resilience system that ensures OpenChrome never hangs and never gives up on element interactions. When one click strategy doesn't work, it automatically escalates through 7 delivery methods, classifies what actually happened after each attempt, uses circuit breakers to prevent wasted time, and learns from success. ### The Problem This Solves Browser automa | Low | 3/20/2026 |
| v1.7.18 | ## v1.7.18 — Cascading Filter: Zero Magic Numbers in AX Resolution This release replaces the score-based AX element resolution with a cascading filter architecture that eliminates all magic numbers and makes element targeting fully deterministic and debuggable. ### Why This Change The v1.7.17 AX-First resolution used a weighted scoring system with **7 hardcoded thresholds** (100, 80, 75, 50, 30, 15 per token, 60 minimum). These numbers were intuitive estimates without empirical validation, cr | Low | 3/20/2026 |
| v1.7.17 | ## v1.7.17 — AX-First Element Resolution: Framework-Agnostic Targeting This release introduces a fundamentally new approach to element targeting that resolves the long-standing failure of interacting with complex UI framework components (Angular Material, React MUI, Vue Vuetify, etc.). Instead of relying solely on CSS-based DOM discovery, OpenChrome now queries the Chrome Accessibility Tree first — the same mechanism screen readers use — which already understands every UI framework natively. # | Low | 3/19/2026 |
| v1.7.16 | ## v1.7.16 — Reliability & Hang Elimination + Prompt Injection Defense This release addresses the **#1 reported user pain point**: operations hanging indefinitely and failing to complete. A comprehensive root cause analysis identified 8 distinct failure modes across the codebase, all of which are fixed in this release. Additionally, a new content sanitization layer provides the first defense against indirect prompt injection attacks. ### Hang Elimination (P0 — Critical) #### Workflow Engine D | Low | 3/18/2026 |
| v1.7.15 | ## v1.7.15 — Hint Engine Hang Prevention This release fixes the root causes behind LLM agent hang loops that the hint engine was failing to prevent in real-world usage, despite having nominal coverage for the error patterns involved. ### Bug Fixes #### Fix A: Missing error-recovery rules (#300) Two common error patterns had no matching hint rule, leaving the LLM without actionable guidance: - **`not editable`** — `form_input` targeting non-editable elements (SVG, div, span) now triggers a hi | Low | 3/12/2026 |
| v1.7.14 | ## v1.7.14 — Critical Bug Fixes for Tool Reliability This release resolves three long-standing issues that caused tool failures and agent hangs during browser automation workflows. ### Bug Fixes #### 1. `javascript_tool` returns empty `{}` for DOM collections (#293, PR #297) **Root Cause:** V8's `returnByValue: true` serialization cannot enumerate `NodeList`, `HTMLCollection`, `DOMTokenList`, `Map`, or `Set` items because they aren't own enumerable properties. This caused `document.querySele | Low | 3/12/2026 |
| v1.7.13 | ## v1.7.13 Release Notes ### New Features #### Enhanced Stealth Navigation Defenses (#290) Significantly expanded anti-detection patches for stealth navigation mode, targeting the most common browser automation fingerprinting vectors beyond `navigator.webdriver`: - **`chrome.runtime` shape spoofing** — Ensures `chrome.runtime` exists with the expected shape, matching real Chrome behavior even without extensions - **Permissions API patching** — Overrides `navigator.permissions.query()` for not | Low | 3/11/2026 |
| v1.7.12 | ## v1.7.12 — Shadow DOM, Token Compression & Stealth Navigation ### 🌗 Shadow DOM Support (#262) Full shadow DOM traversal across the entire tool suite — elements inside shadow roots (open and closed) are now discoverable, clickable, and inspectable. - **DOM Serializer**: Traverses open/closed shadow roots with `--shadow-root-- (type)` separator rendering, respects `maxDepth` and `interactiveOnly` filters - **Element Discovery**: New shared `element-discovery.ts` module with 3-phase search — | Low | 3/9/2026 |
| v1.7.11 | ## What's New ### SPA Readiness — Navigate now waits for framework rendering (#258) `navigate` previously returned as soon as `domcontentloaded` fired, before SPA frameworks (React, Vue, Angular, Swagger UI) had rendered their content. This caused `find`, `wait_for`, and `read_page` to operate on a near-empty DOM — for example, seeing only 9 elements on a Swagger page that should have hundreds. **v1.7.11 fixes this with three changes:** ### 1. DOM Stability Wait (`smart-goto.ts`) — PR #259 | Low | 3/8/2026 |
| v1.7.10 | ## v1.7.10 — Error Handling & Resilience This release fixes structural error handling issues that caused cascade failures in parallel tool calls, improved timeout detection, and better form field matching. ### P0: Session Init Structural Fix (#254) Session initialization (`getOrCreateSession` + `puppeteer.connect()`) was running **outside** the tool-level try/catch in `handleToolsCall()`. When session init timed out, the error propagated as a JSON-RPC `-32603` protocol error instead of an `MC | Low | 3/8/2026 |
| v1.7.9 | ## v1.7.9 — Chrome 136+ Compatibility & Windows Lock Detection This release resolves the critical "Chrome debug port 9222 not available after 60000ms" issue reported in #246, along with related profile locking and anti-automation improvements. ### Bug Fixes #### Always Use Persistent Profile for Auto-Launched Chrome (#248) Chrome 136+ rejects `--remote-debugging-port` when the default `--user-data-dir` is used. OpenChrome now always routes to its persistent profile (`~/.openchrome/profile/`) | Low | 3/8/2026 |
| v1.7.8 | ## v1.7.8 — Navigation Timeout Graceful Recovery ### Bug Fix #### Navigation Timeout No Longer Stalls LLM Workflow Navigation timeouts ("Navigation timeout of 30000 ms exceeded") previously returned `isError: true`, causing LLMs to halt and wait for user input. This commonly occurs when: - A local dev server (`localhost:3000`) hasn't started yet - A page is slow to load (heavy SPA, large assets) - A server is temporarily unresponsive **Before**: LLM stalls → user must type "continue" **Af | Low | 3/7/2026 |
| v1.7.7 | ## v1.7.7 — Graceful Error Recovery & Chrome Profile Selection ### Highlights This release significantly improves LLM workflow continuity by fixing three categories of errors that caused unnecessary interruptions, and adds Chrome profile selection support. --- ### New Features #### Chrome Profile Selection (`--profile-directory`) - **CLI flag**: `--profile-directory "Profile 1"` to launch Chrome with a specific user profile - **Environment variable**: `CHROME_PROFILE_DIRECTORY` for MCP serv | Low | 3/7/2026 |
| v1.7.6 | ## v1.7.6 — Hint Escalation False Positive Fixes This release fixes a critical issue where the hint escalation system would fire `🛑 CRITICAL — you MUST change approach` on **legitimate batch operations** like opening 20 tabs in parallel, filling multi-field forms, or reading multiple pages. The root cause was that advisory optimization hints shared the same unbounded escalation ladder as stuck-detection hints. ### What was broken The escalation system (#71, v1.7.3) escalates hints based on p | Low | 3/6/2026 |
| v1.7.5 | ## v1.7.5 Release Notes This release improves OpenChrome's handling of CAPTCHAs, browser storage synchronization, and hint engine accuracy. It reflects OpenChrome's core design philosophy: **extending the user's authenticated browser session** rather than bypassing security measures. --- ### 🛡️ CAPTCHA Human-in-the-Loop Flow (#221, PR #226) OpenChrome now provides **immediate guidance** when encountering CAPTCHAs, bot-checks, or access-denied pages — instead of wasting 3-5 tool calls before | Low | 3/6/2026 |
| v1.7.4 | ## v1.7.4 — Stale Ref Auto-Recovery & Parallel Worker Isolation This release completes the session stability work started in v1.7.3, focusing on three remaining gaps in the ref lifecycle and adding enriched error responses for LLM self-recovery. ### Highlights - **Zero stale refs after Chrome restart** — All `backendDOMNodeId` values are now properly invalidated when Chrome restarts or tabs are closed externally - **Parallel worker isolation** — Workers no longer poison each other's progress | Low | 3/6/2026 |
| v1.7.3 | ## Session Stability & Error Recovery (Issue #205) This release fixes three classes of errors that caused browser automation sessions to stall: ### Fixes **Ref namespace collision (PR #207)** - `query_dom` now uses `el_N` labels instead of `ref_N`, eliminating collision with `RefIdManager`'s `ref_N` namespace - Prevents `form_input` and `computer` from operating on wrong elements **Stale session state after Chrome restart (PR #208)** - `SessionManager` now listens for `reconnect_failed` even | Low | 3/6/2026 |
| v1.7.2 | ## v1.7.2 — Session Stability & Auto-Recovery ### 🔧 Critical Fix: WebSocket Timeout Auto-Recovery **The most impactful change in this release.** Previously, when `puppeteer.connect()` timed out (commonly after macOS sleep/wake), the MCP server would fail permanently because `isConnectionError()` didn't recognize the timeout as a recoverable connection error. The auto-reconnect path (`forceReconnect()`) was effectively dead code for this scenario. **Now fixed:** Two new patterns added to `isC | Low | 3/6/2026 |
| v1.7.1 | ## v1.7.1 — Universal `expand_tools` Compatibility ### Problem OpenChrome uses a progressive tool disclosure system (Tier 1 → Tier 2 → Tier 3) to avoid overwhelming AI agents with 46 tools at once. When an agent calls `expand_tools`, the server expands the tier and sends a `notifications/tools/list_changed` notification per the MCP spec, prompting the client to re-fetch the tool list. However, **most MCP clients do not support `notifications/tools/list_changed`**. This includes: - **Google A | Low | 3/6/2026 |
| v1.6.12 | ## CDP Timeout Resilience This release resolves the `Runtime.callFunctionOn timed out` error that caused tools like `lightweight_scroll` to hang indefinitely when page dialogs or unresponsive scripts blocked CDP calls. ### What's Changed **Per-call timeout protection for all 18 tools** (#182) Every `page.evaluate()` call across all tool files is now wrapped with `withTimeout()`, providing labeled per-call timeouts (2-15s based on operation complexity). Previously, only 3 of 18 tools had this | Low | 3/5/2026 |
| v1.6.11 | ## What's Changed * feat: improve tool reliability with polling, diagnostics, and scroll fixes by @shaun0927 in https://github.com/shaun0927/openchrome/pull/181 **Full Changelog**: https://github.com/shaun0927/openchrome/compare/v1.6.10...v1.6.11 | Low | 3/5/2026 |
| v1.6.10 | ## What's Changed * fix: add remaining anti-hang defenses (renderer crash, print dialog, GPU, downloads) by @shaun0927 in https://github.com/shaun0927/openchrome/pull/179 * fix: Gemini API compatibility — use string type for expand_tools tier enum by @shaun0927 in https://github.com/shaun0927/openchrome/pull/180 **Full Changelog**: https://github.com/shaun0927/openchrome/compare/v1.6.9...v1.6.10 | Low | 3/4/2026 |
| v1.6.9 | ## v1.6.9 — Anti-Hang: Never-Stop Browser Automation This release focuses on **eliminating all known causes of browser tab freezes and hangs** during automated operations. Previously, JavaScript dialogs (alert/confirm/prompt/beforeunload) could freeze a tab indefinitely, blocking all subsequent CDP commands. This release adds comprehensive defenses at three layers. ### Highlights - **JavaScript dialogs can no longer freeze tabs** — all dialogs are auto-dismissed instantly - **Screenshot opera | Low | 3/3/2026 |
| v1.6.8 | ## What's Changed * Extract duplicated magic numbers and fix tool naming inconsistencies by @shaun0927 in https://github.com/shaun0927/openchrome/pull/170 * fix: prevent duplicate Chrome instances on slow launch (fixes #171) by @shaun0927 in https://github.com/shaun0927/openchrome/pull/172 **Full Changelog**: https://github.com/shaun0927/openchrome/compare/v1.6.7...v1.6.8 | Low | 3/3/2026 |
| v1.6.7 | ## What's Changed * feat: add ProgressTracker class for LLM wandering prevention by @shaun0927 in https://github.com/shaun0927/openchrome/pull/166 * feat: integrate ProgressTracker into HintEngine at priority 50 by @shaun0927 in https://github.com/shaun0927/openchrome/pull/168 * feat: auth redirect fail-fast and enhanced login detection by @shaun0927 in https://github.com/shaun0927/openchrome/pull/167 * benchmark: add Twitter profile scraping benchmark suite by @shaun0927 in https://github.com/s | Low | 3/3/2026 |
| v1.6.6 | ## What's Changed * fix: stop Chrome from auto-reopening when user closes it by @shaun0927 in https://github.com/shaun0927/openchrome/pull/160 * refactor: consolidate click_element and wait_and_click into interact tool by @shaun0927 in https://github.com/shaun0927/openchrome/pull/161 * feat: add tiered tool exposure system for reduced cognitive load by @shaun0927 in https://github.com/shaun0927/openchrome/pull/162 **Full Changelog**: https://github.com/shaun0927/openchrome/compare/v1.6.5...v1. | Low | 3/2/2026 |
Dependencies & License Audit
Loading dependencies...
Similar Packages
sidebuttonAI agent platform and MCP server — knowledge packs, AI agent tools, workflow automation, and CLI1.1.1