New: 🧠 AI Vector Search • Semantic Clustering • Mock APIs • 24 Datasets

Docs · MCP

MCP integration

Connect Claude Code to your test data using the Model Context Protocol (MCP). Five consolidated tools (down from 24) with an efficient router pattern, 24 queryable datasets, and built-in mock API management.

What's New in v2.1

🧠 AI Vector Search - Semantic search & failure clustering using Jina v3 embeddings + Cohere reranking
24 queryable datasets - Including clustered_failures, semantic_search, mock APIs, org_suites, suite_tests, inactive_tests, and setup_guide
find_similar action - Find similar failures across current or historical runs
5 tools instead of 24 - Router pattern with ~83% token reduction
HTTP Streamable transport - More reliable than legacy SSE
Semantic definitions - Metric formulas prevent AI hallucinations

Installation

1Quick Start (OAuth - Recommended)

One command to connect Claude Code to Exolar : authenticates via browser:

claude mcp add exolar-qa --transport http https://exolar.agentical.work/api/mcp/mcp -s user

When prompted, select "Authenticate" → browser opens → log in → done!

Verify installation: Run claude mcp list to see "exolar-qa" connected

2Alternative: Manual Token

If OAuth doesn't work, get a token from /settings/mcp:

claude mcp add exolar-qa --transport http https://exolar.agentical.work/api/mcp/mcp -s user \
  --header "Authorization: Bearer YOUR_TOKEN_HERE"

🤖AI-Guided CI/CD Setup

After connecting, ask Claude Code:

"Help me integrate Exolar with my Playwright tests"

Claude uses an Integration Engineer persona to guide you through GitHub Actions setup with step-by-step instructions.

Architecture: Router Pattern

Instead of 24 individual tools, the MCP server uses a two-level router pattern:

Level 1: Tool

Which category?

• explore (discovery)
• query (data retrieval)
• action (heavy ops)
• definition (semantics)
• config (setup)

Level 2: Dataset/Action

Which specific data?

• executions, failures
• flaky_tests, trends
• mock_interfaces, mock_routes
• org_suites, suite_tests
• compare, classify, reembed
• metric formulas
• CI/CD snippets

Benefits

83% fewer tokens - Tool definitions reduced from ~3,000 to ~500 tokens
Unified filters - Consistent interface across all datasets
Better discovery - explore_exolar_index shows what's available
Flexible routing - Easy to add new datasets

Available Tools (5 Total)

All tools support organization-scoped access with automatic filtering.

discoveryexplore_exolar_index

Discovery tool - call FIRST to learn what data exists. Lists datasets, branches, suites, or metrics with optional search filtering.

queryquery_exolar_data

Universal data retrieval router. Retrieves data from any of the 24 available datasets using a unified filter interface.

actionperform_exolar_action

Execute heavy operations: compare executions, generate reports, classify failures, find similar failures, re-embed tests, or manage mock APIs.

metadataget_semantic_definition

Get metric definitions to prevent AI hallucinations. Returns formula, thresholds, unit, and related tools for any metric.

metadataget_installation_config

Get CI/CD integration code snippets and setup instructions. Returns ready-to-use config for Playwright reporter, GitHub Actions, and environment variables.

Available Datasets (24 Total)

Use query_exolar_data with these datasets:

executions

List test executions with optional filters (branch, suite, status, dates)

execution_details

Full execution data with all test results and artifacts (requires execution_id)

failures

Failed tests with AI-enriched context and error details

flaky_tests

Tests with flakiness history (min_runs, include_resolved filters)

trends

Time-series metrics with flexible granularity (hour, day, week, month)

dashboard_stats

Overall metrics summary (total executions, pass rate, avg duration)

error_analysis

Error type distribution breakdown

test_search

Search tests by name or file path (requires query param)

test_history

Execution history for a specific test (requires test_signature)

flakiness_summary

Overall flakiness metrics and worst offenders

reliability_score

Suite health score (0-100) with breakdown and trend

performance_regressions

Tests slower than historical baseline

execution_summary

Aggregated execution overview without full test list

execution_failures

Only failed tests from an execution with error grouping

clustered_failures

AI-grouped failures by similarity (requires execution_id, reduces 50+ failures to root causes)

semantic_search

Natural language search for failures using vector embeddings (requires query)

setup_guide

CI/CD integration setup instructions filtered by provider and framework

org_suites

All test suites for the organization with pass rates and last-run stats

suite_tests

Individual tests within a suite (requires suite filter)

inactive_tests

Tests that haven't run recently : useful for test suite cleanup

mock_interfaces

List mock API interfaces created for the organization

mock_routes

Routes configured on a mock interface (requires interface_id)

mock_rules

Response rules for a mock route (requires route_id)

mock_logs

Request logs for a mock interface with matched rule details

Migration from v1.0

If you were using the previous 24-tool version, here's how to migrate:

Old Tool	New Tool	Parameters
`list_branches`	`explore_exolar_index`	`{ category: "branches" }`
`get_executions`	`query_exolar_data`	`{ dataset: "executions" }`
`get_flaky_tests`	`query_exolar_data`	`{ dataset: "flaky_tests" }`
`compare_executions`	`perform_exolar_action`	`{ action: "compare" }`
See full mapping in documentation

Usage Examples

After connecting, you can ask Claude things like:

“Show me what datasets are available”

“List all branches with their stats”

“What metrics can I query?”

“Get recent test failures on main branch”

“Show me the flakiest tests with at least 5 runs”

“What were the test results from the last CI run?”

“Search for tests related to login”

“Get metrics for the past week”

“What's the health score for my test suite?”

“Are there any performance regressions on the main branch?”

“Show me tests that got slower in the last 24 hours”

“What's the error distribution for failed tests?”

“Cluster the failures from the last run by root cause”

“Find tests with timeout errors using semantic search”

“Show me similar failures to this test”

“Group 50+ failures into meaningful clusters”

“Compare the last two runs on main branch”

“What tests broke in the feature branch compared to main?”

“Show me new failures between main and feature-auth”

“Is this test failure a flake or a real bug?”

“Classify the failure in test 'should login successfully'”

“Create a mock API for the user service”

“List all mock interfaces for my org”

“Add a route that returns 404 for unknown users”

“Help me set up MCP integration”

“How do I configure GitHub Actions?”

“What's the formula for pass rate?”

“How is reliability score calculated?”

Conversational CI/CD Setup

The get_installation_config tool uses an Integration Engineer persona that guides you through setup step-by-step.

🔍

Discovery

Claude asks about your CI provider (GitHub Actions recommended), monorepo structure, and existing tests

🎯

Adaptation

Receives CI-specific instructions with exact secrets management steps for GitHub Actions

✅

Validation

Gets dry run commands to verify integration before pushing to CI

Example Conversation

You: “Help me integrate Exolar with my Playwright tests”

Claude (Integration Engineer): “I'll help you set up Exolar! First, which CI provider are you using? (GitHub Actions recommended, or running locally?)”

You: “GitHub Actions”

Claude: “Perfect! Here are your GitHub Actions setup instructions: 1. Add this secret in Settings > Secrets > Actions: Name: EXOLAR_API_KEY Value: [from /settings/mcp] 2. Install the package: npm install @exolar/reporter 3. MERGE (don't replace!) the reporter into your playwright.config.ts reporters array 4. Update your .github/workflows/playwright.yml with... Would you like me to explain how to merge the reporter config?”

Key Benefits

• No guessing where to put the EXOLAR_API_KEY
• Clear instructions: MERGE vs REPLACE config files
• GitHub Actions-specific secrets management path
• Monorepo-aware guidance (where to place reporters, per-package config)
• Built-in validation commands (dry run locally before CI push)

Router Pattern Alternative

You can also use the router pattern to get pre-filtered configuration programmatically:

query_exolar_data({
  dataset: "setup_guide",
  filters: {
    ci_provider: "github",  // or "local" (v2.1)
    framework: "playwright",
    monorepo: false,        // or true
    section: "all"          // or specific section
  }
})

Output Formats

Markdown (Default)

• CLI-friendly tables
• ~70% fewer tokens than JSON
• Human-readable
• Easier to scan

JSON

• Structured data
• Programmatic parsing
• Full details
• API integration

Security

JWT tokens verified using Neon Auth JWKS
All data automatically scoped to your organization
Database-level RLS policies for additional protection
HTTPS-only communication (encrypted in transit)
No cross-tenant access - users only see their org's data

Troubleshooting

“Invalid or expired token”

Visit /settings/mcp to generate a new token. Tokens expire after 30 days.

“User not found”

Ensure you're logged into the dashboard and have an organization assigned. Generate a new token from /settings/mcp after logging in.

“Tool not found”

You may be using old tool names. Use the 5 consolidated tools: explore_exolar_index, query_exolar_data, perform_exolar_action, get_semantic_definition, get_installation_config.

Connection fails

Check: (1) Dashboard is accessible, (2) NEON_AUTH_JWKS_URL is set in Vercel, (3) URL is correct (/api/mcp/mcp not /api/mcp).

Additional Resources

• Settings Page - Get your configuration
• Claude Code Docs - Official Claude Code documentation
• MCP Specification - Model Context Protocol spec

MCP integration

What's New in v2.1

Installation

1Quick Start (OAuth - Recommended)

2Alternative: Manual Token

🤖AI-Guided CI/CD Setup

Architecture: Router Pattern

Benefits

Available Tools (5 Total)

Parameters (3)

Response Fields

Example Usage

Parameters (4)

Response Fields

Example Usage

Parameters (2)

Response Fields

Example Usage

Parameters (1)

Response Fields

Example Usage

Parameters (1)

Response Fields

Example Usage

Available Datasets (24 Total)

Migration from v1.0

Usage Examples

Conversational CI/CD Setup

Discovery

Adaptation

Validation

Example Conversation

Key Benefits

Router Pattern Alternative

Output Formats

Markdown (Default)

JSON

Security

Troubleshooting

“Invalid or expired token”

“User not found”

“Tool not found”

Connection fails

Additional Resources