What's New
Latest updates, features, and improvements to Exolar QA
🧠 AI Vector Search - Smart Failure Clustering & Semantic Search
Transform test failure triage from hours to minutes with AI-powered clustering and semantic search. Using Jina v3 embeddings and Cohere reranking, Exolar now automatically groups 50+ failures into root cause clusters and lets you search failures using natural language.
What's Included
Smart Failure Clustering
Automatically group 50+ failures into root cause clusters. See '50 failures → 3 issues' instead of scrolling through repetitive errors. Toggle between list and clustered views in the execution details page.
Learn more →Semantic Test Search
Find tests by intent, not just file names. Search 'timeout errors' or 'login failures' using natural language. Three search modes: Hybrid (default), AI (semantic), and Keyword.
Learn more →Find Similar Failures
New MCP action 'find_similar' finds related failures using vector similarity. Scope to current execution or search across historical runs to identify recurring issues.
Learn more →Jina v3 + Cohere Reranking
512-dim Matryoshka embeddings via Jina v3 with asymmetric query/passage support. Two-stage retrieval with Cohere rerank-english-v3.0 for precision. Automatic fallback to Gemini if Jina unavailable.
Learn more →2 New MCP Datasets
Query clustered_failures to get AI-grouped failures by similarity. Use semantic_search for natural language test search. Now 16 total datasets via query_exolar_data.
Learn more →AI Insights Metrics
New metric category with cluster_reduction (grouping efficiency), similarity_score (vector similarity), embedding_coverage (failures with embeddings), and search_relevance scores.
Learn more →AI Insights Dashboard Card
New dashboard card showing embedding coverage, indexed failures, and cluster reduction stats at a glance.
Learn more →Dedicated AI Search Page
Full-page semantic search at /dashboard/search with mode selector, results table, similarity scores, and CSV export.
Learn more →Similar Failures in Test History
Test History modal now includes a 'Similar Issues' tab showing related failures from the last 30 days with match percentages.
OAuth 2.1 Authentication - No Token Copying Needed
Seamless MCP authentication via browser login. Simply run the add command and authenticate in your browser — no more copying tokens from the dashboard. Uses OAuth 2.1 with PKCE for security.
What's Included
One-Command Setup
Run 'claude mcp add --transport http exolar-qa https://exolar.ai-innovation.site/api/mcp/mcp' and authenticate via browser. Done!
Learn more →Browser-Based Login
When you add the MCP server, Claude Code automatically opens your browser to log in. After authentication, you're connected — no tokens to copy or paste.
Learn more →OAuth 2.1 with PKCE
Industry-standard security using OAuth 2.1 Authorization Code flow with PKCE (Proof Key for Code Exchange). Tokens are securely exchanged without exposure.
Learn more →30-Day Token Validity
Access tokens are valid for 30 days. When they expire, simply re-authenticate via the same browser flow — no manual token management required.
Learn more →AI-Guided CI/CD Setup with Integration Engineer Persona
Transform CI/CD setup from confusing config dumps to conversational pair programming. Claude Code now adopts an Integration Engineer persona that asks about your environment first, then provides tailored instructions for your specific CI provider and project structure.
What's Included
Conversational Setup Experience
Claude asks discovery questions (CI provider, monorepo structure) before providing configuration. No more guessing where to put tokens or how to merge config files.
Learn more →GitHub Actions-Specific Instructions
Get GitHub Actions-specific instructions with exact secrets management steps (Settings > Secrets > Actions path). Currently focused on GitHub Actions with more CI providers coming soon.
Learn more →Monorepo-Aware Guidance
Claude explicitly asks if you're using a monorepo and provides guidance on where to place reporters, how to configure per-package Playwright configs, and manage environment variables.
Learn more →Built-in Validation
Claude suggests dry run commands to verify integration locally before pushing to CI. Catch configuration issues early with 'npx playwright test --reporter=@exolar/reporter'.
Learn more →Contextual Troubleshooting
Get instant help for common issues: 401 Unauthorized (token expiration), missing data (reporter not in config), module errors (missing npm install). Claude provides exact fix steps based on your error.
Learn more →Setup Guide Dataset
New 'setup_guide' dataset in query_exolar_data allows programmatic access to filtered configuration with ci_provider, framework, monorepo, and section filters.
Learn more →MCP Integration v2.0 - Router Pattern & Semantic Layer
Major update to MCP integration introducing a consolidated 5-tool router pattern, semantic definitions for AI safety, and HTTP Streamable transport. This update reduces token overhead by 83% and makes Claude Code integration more reliable.
What's Included
5-Tool Router Pattern
Consolidated from 24 individual tools to 5 consolidated tools (explore, query, action, definition, installation). This architectural change reduces token overhead from ~3,000 to ~500 tokens - an 83% reduction.
Learn more →14 Queryable Datasets
Universal data access through the query_exolar_data tool. Datasets include: executions, failures, flaky tests, trends, dashboard stats, error analysis, test search, test history, flakiness summary, reliability score, performance regressions, and more.
Learn more →Semantic Definitions Layer
New get_semantic_definition tool provides explicit metric definitions with formulas, thresholds, and units. This prevents AI hallucinations by giving Claude precise information about what each metric means and how it's calculated.
Learn more →HTTP Streamable Transport
Upgraded to the new MCP standard transport protocol (HTTP Streamable). Replaces legacy SSE with automatic retry logic, better payload chunking, and improved reliability. Uses Vercel's mcp-handler v1.0.7.
Learn more →Dual Authentication
Enhanced security with dual-token validation: Neon Auth tokens (RS256 with JWKS verification) for browser sessions and MCP tokens (HS256) for API access. Organization-scoped queries with RLS protection at the database level.
Learn more →Simplified Integration
One-click MCP token generation at /settings/mcp. No complex setup required - just paste the configuration URL into Claude Code and start querying your test data immediately.
Learn more →⚠️Migration from v1.x
- •BREAKING: Old tool names (get_executions, get_flaky_tests, etc.) no longer work. Use the migration table in the documentation to update your prompts to the new 5-tool pattern.
- •If using custom MCP clients: Update transport to HTTP Streamable (endpoint: /api/mcp/mcp)
- •Regenerate your MCP token at /settings/mcp for the latest security improvements
- •Review the semantic definitions to understand exact metric calculations
Launched January 10, 2026