ProvaLab MCP — Testing Infrastructure for Claude Code

ProvaLab exposes 71 testing tools over the Model Context Protocol (MCP). Point your Claude Code (or any MCP-capable agent) at our endpoint and your AI assistant can drive real browser sessions, run security scans, execute API tests, and manage test cases — all with your own LLM subscription paying for the chat side, and ProvaLab charging only for the infrastructure (browsers, scans, device time, storage) that actually runs.

Who this is for

• You use Claude Code, Claude Desktop, Cursor, Zed, or any MCP-capable agent
• You want your AI assistant to actually run tests, not just talk about them
• You'd rather pay a few cents per browser-minute than a flat SaaS fee for AI you're already paying for elsewhere

5-minute install

Claude Code (recommended)

1. Install the auth helper (one-time, per machine).

The plugin needs a short-lived OAuth token for every MCP call. The provalab-auth CLI handles the device-flow login + silent refresh.

macOS / Linux:

curl -sSf https://plugins.provalab.io/install.sh | sh
provalab-auth login

Windows (PowerShell):

irm https://plugins.provalab.io/install.ps1 | iex
provalab-auth login

provalab-auth login opens a browser (or prints a code + URL for headless machines), you sign in to ProvaLab, and the token is cached at ~/.config/provalab/auth.json.

2. Install the plugin.

/plugin marketplace add provalab-dev/claude-code-plugin
/plugin install provalab@provalab-marketplace

Restart Claude Code, then run /mcp — the provalab server should show as connected. You now have the ProvaLab tool list plus /provalab-test, /provalab-status, /provalab-results slash commands and the run/triage skills.

Direct MCP URL (Claude Desktop, Cursor, Zed, custom clients)

ProvaLab's MCP endpoint speaks the streamable-HTTP transport. Point your client at the URL below — it handles OAuth automatically on first use, so no provalab-auth CLI needed.

claude mcp add provalab https://mcp.provalab.io/mcp

Or in your client's MCP config:

{
  "mcpServers": {
    "provalab": {
      "type": "http",
      "url": "https://mcp.provalab.io/mcp"
    }
  }
}

Don't use /mcp/sse with "type": "sse". The legacy SSE transport is still mounted but is deprecated and connections time out after 30s. Always use /mcp with "type": "http".

What you can ask Claude once installed

The tools group into four testing pillars + one data layer.

---

Pillar 1 — UI Testing

Run real browser automation on ProvaLab's cloud infrastructure. Chrome, Firefox, Edge, mobile emulators, device cloud.

Tools

Tool	What it does
execute_pytest	Run a pytest suite against a target. Returns pass/fail + logs + screenshots
provision_browser	Spin up a headless/VNC-visible browser with a chosen version
create_browser_session	Open an interactive browser session (persistent, user can watch)
close_browser_session	Tear down a session + reclaim resources
authenticate_session	Log into an app via OTP from a real device (SMS/email)
crawl_site	Walk a site's routes to discover pages
save_test_artifacts	Attach screenshots, videos, HAR files to a run
get_session_artifacts	Retrieve screenshots / videos / logs from a session
get_execution_result	Fetch the structured result of a prior run
analyze_page	Extract elements / selectors / accessibility hints from a page

Example prompts

"Open https://staging.example.com in Chrome and screenshot the pricing page."

"Run the smoke suite I have in pytest-tms format against staging, then show me any failures with screenshots."

"Log into my app at app.example.com using my test credentials, grab the OTP from my phone, then verify the dashboard loads."

Infrastructure consumed (what you pay for)

• Browser session minutes
• Mobile emulator / real-device time
• Video recording GB-months
• Screenshot & artifact storage

---

Pillar 2 — API Testing

Execute HTTP requests, inspect responses, maintain a catalog of your APIs.

Tools

Tool	What it does
execute_api_request	Fire a one-shot HTTP request with full headers/body/auth. Returns full response + timing
get_api_catalog	List the APIs your project has registered
get_api_workbench	Fetch saved request collections (Postman-style)

Example prompts

"POST a login request to api.example.com/auth with these credentials, and show me the full response headers."

"Run the entire users collection against staging and tell me which endpoints return non-2xx."

"Hit /health on all three environments and compare the response times."

Infrastructure consumed

• API requests per month
• Request/response storage

Coming soon

• api_contract_validate — OpenAPI spec vs. live-service drift detection
• api_regression — compare current response vs. recorded baseline
• api_mock_server — spin up a mock service from a spec

---

Pillar 3 — Security Testing

OWASP ZAP-backed scanning with active and passive modes, fuzz testing, formatted reports.

Tools

Tool	What it does
security_scan	Run a baseline or active ZAP scan against a URL. Returns findings with severity
security_fuzz	Fuzz endpoints with injection / XSS / auth-bypass payloads
security_report	Produce a human-readable security report (PDF/Markdown) from a scan

Example prompts

"Run the 'Nightly DAST' scan for project 7 and wait for the gate."

"Fuzz the /api/users/:id endpoint for IDOR and SQL injection."

"Give me a security report for scan ID 42 — I need to share it with legal."

Infrastructure consumed

• Scan minutes (ZAP container runtime)
• Report retention

Coming soon

• security_sast — static analysis (semgrep / bandit) on your source code
• security_deps — dependency vulnerability scan (Trivy / Snyk-equivalent)
• security_secrets — leaked secrets scan on repo
• security_container — container image scan

---

Pillar 4 — Performance Testing

Coming soon. We're building k6-backed load, stress, spike, and endurance testing. Target MCP tools:

• perf_load_test — sustained load at N vUsers for M minutes
• perf_stress_test — ramp vUsers until the service breaks
• perf_spike_test — sudden traffic spike recovery
• perf_endurance_test — long soak test for memory leaks
• perf_profile_page_load — Lighthouse scores for page load
• perf_compare — regression vs. a baseline run

Status: Design phase. Release updates are posted on provalab.io and in the plugin catalogue.

---

Data layer — TMS (Test Management System)

39 CRUD tools over the ProvaLab TMS. Lets Claude manage projects, versions, modules, test cases, plans, runs, defects — all natively, no code written.

Read tools (list & inspect)

Tool	What it does
tms_list_projects	List projects in your organization
tms_get_project	Details for one project
tms_get_project_stats	Roll-up counts (modules, suites, cases, runs, features) for a project
tms_list_project_members	Users on a project + their roles
tms_list_versions	Project versions / releases
tms_get_version	Details for one version (e.g. "8.0.0")
tms_list_modules	Modules in a project
tms_list_test_suites	Suites in a project or module
tms_list_test_cases	Test cases (filterable by priority, type, status)
tms_list_test_runs	Test runs (filterable by status, environment)
tms_get_test_run_results	Pass/fail results for a specific run
tms_list_test_plans	Test plans
tms_get_test_plan	Details for one plan
tms_get_plan_coverage	Coverage breakdown for a plan
tms_list_requirements	Requirements
tms_get_requirement_coverage	Which requirements have tests, which don't
tms_list_features	Features
tms_list_issues	Issues / defects
tms_get_issue	Details for one issue
tms_search	Free-text search across TMS entities

Write tools (require approval in interactive mode)

Tool	What it does
tms_create_project	New project
tms_create_version	New version/release
tms_create_module	New module
tms_create_test_suite	New suite
tms_create_test_case	New test case
tms_update_test_case	Edit an existing case
tms_create_test_plan	New test plan
tms_create_test_run	New run (attached to a plan)
tms_report_result	Report a pass/fail for a case in a run
tms_create_requirement	New requirement
tms_update_requirement	Edit requirement
tms_delete_requirement	Delete requirement
tms_create_feature	New feature
tms_update_feature	Edit feature
tms_delete_feature	Delete feature
tms_create_issue	File a bug
tms_update_issue	Update an issue
tms_link_test_case_to_requirement	Trace case → requirement
tms_link_test_case_to_feature	Trace case → feature

Example prompts

"List all projects, then show me the one with the most open bugs."

"For project ForceFlow, create a new version 9.0.0, draft the regression test plan for it, and link every HIGH-priority requirement to the plan."

"Which test cases in Login module have no automation script? File those as 'needs automation' bugs."

Infrastructure consumed

• TMS CRUD operations are free during the beta period.

---

Supporting tools (invisible but useful)

Claude will use these automatically when a prompt needs them — you don't call them directly.

Knowledge & context:

• get_project_context, get_app_profile, get_tms_schema, get_domain_history

Memory:

• save_learning, get_learnings — persist what Claude figures out about your app
• save_working_context, get_working_context — maintain state across sessions

Workspace:

• list_workspace_files, read_workspace_file — read files you've uploaded
• get_ui_test_inventory — list test files Claude authored

Human-in-the-loop:

• request_user_input — Claude pauses to ask you for approval before destructive actions

Sync:

• publish_to_master, get_knowledge_gaps, get_session_history

---

Current pricing (beta)

Pillar	Status	Price
UI Testing	Live	Free during beta. Fair-use session minutes, premium tier launching Q1
API Testing	Live	Free during beta
Security Testing	Live	Free during beta. Per-scan pricing launching Q1
Performance Testing	Coming soon	TBD
TMS CRUD	Live	Free, always (bring your own LLM)

No charge for LLM tokens. Your Claude Code / Max / Pro subscription (or any MCP-capable agent) pays for the AI side. ProvaLab charges only for the infrastructure that actually runs tests.

Next steps

• Install → /plugin marketplace add provalab-dev/claude-code-plugin then /plugin install provalab@provalab-marketplace in Claude Code
• First prompt to try → "List my ProvaLab projects"
• Report issues → github.com/provalab-dev/claude-code-plugin/issues
• Pricing updates → provalab.io/pricing

Release updates and new tools are announced on provalab.io and in the plugin catalogue.