LvcidPsyche/auto-browser

v1.2.0 Feature

This release adds 2 notable features for engineering teams evaluating rollout.

Published 1mo MCP Browser & Automation

✓ No known CVEs patched

✓ No known CVEs patched in this version

Topics

ai-agent ai-agents browser-automation claude docker fastapi

+6 more

llm local-first mcp novnc playwright self-hosted

Affected surfaces

auth deps

Summary

AI summary

Verifiable Witness receipts enable on‑demand chain verification and sturdier session isolation adds configurable resource caps.

Type	Severity	Summary	CVE
Feature
Feature	Low	Adds verifiable witness receipt verification endpoint. Adds verifiable witness receipt verification endpoint. Source: llm_adapter@2026-06-10 Confidence: high	—
Feature	Low	Adds per-session resource caps for isolated browsers (memory, PID, CPU). Adds per-session resource caps for isolated browsers (memory, PID, CPU). Source: llm_adapter@2026-06-10 Confidence: high	—
Feature	Low	Unifies user‑agent pool to current Chrome 149 strings for all layers. Unifies user‑agent pool to current Chrome 149 strings for all layers. Source: llm_adapter@2026-06-10 Confidence: high	—
Dependency	Low	Updates Playwright to 1.60, uvicorn to 0.49, APScheduler to 3.11.2. Updates Playwright to 1.60, uvicorn to 0.49, APScheduler to 3.11.2. Source: llm_adapter@2026-06-10 Confidence: high	—
Bugfix	Low	Logs exceptions with context in cleanup/capture paths. Logs exceptions with context in cleanup/capture paths. Source: llm_adapter@2026-06-10 Confidence: high	—

Full changelog

Verifiable Witness receipts. Receipts were always hash-chained at write time; now you can check the chain on demand. GET /sessions/{id}/witness/verify and the read-only browser.verify_witness MCP tool (curated profile) walk the full chain and report the first divergent receipt if the log was altered, reordered, truncated, or unparseable.
Sturdier session isolation. Per-session browser containers now get resource caps (ISOLATED_BROWSER_MEM_LIMIT 4g default, ISOLATED_BROWSER_PIDS_LIMIT 2048, ISOLATED_BROWSER_CPUS opt-in), and startup reaps containers orphaned by a crashed controller.
Fresh dependency floor. Playwright 1.56 → 1.60 on controller and browser-node together (WS protocol kept aligned), uvicorn 0.49, APScheduler 3.11.2; dev floors raised. Clears the open Dependabot batch.
Unified, current user-agent pool. One authoritative CHROME_UA_POOL (Chrome 149, June 2026 stable) now feeds both the fingerprint layer and the USER_AGENT_POOL default, which previously shipped stale Chrome 122–124 strings. Firefox/Safari UAs no longer cycle on the Chromium engine.
Quieter failure modes. Cleanup/capture paths that swallowed exceptions now log with context.
Security reporting channel. SECURITY.md now points at GitHub private vulnerability reporting.
CI host tests now run a Python 3.11 + 3.14 matrix; coverage gate holds at 80% (81% actual).

Full details in CHANGELOG.md.

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 LvcidPsyche/auto-browser

Get notified when new releases ship.

About LvcidPsyche/auto-browser

Give your AI agent a real browser — with a human in the loop. Open-source MCP-native browser agent.