Beta: Provar MCP is currently in Beta. This is offered to all Provar users at no additional cost, and is an open source project hosted on GitHub here. General Availability is coming soon. We welcome feedback via GitHub Issues.

Provar MCP is an AI-assisted quality layer built directly into the Provar DX CLI. It implements the Model Context Protocol (MCP) — an open standard that lets AI assistants call tools on your behalf — and exposes a rich set of Provar project operations to AI clients such as Claude Desktop, Claude Code, and Cursor.

Once connected, your AI assistant can:

• Inspect your Provar Automation project and surface coverage gaps
• Generate Java Page Objects and XML test case skeletons
• Validate every level of the test hierarchy (test cases, suites, plans, and the full project) against 30+ quality rules
• Set up and manage your provardx-properties.json run configuration
• Trigger Provar Automation test runs and Provar Quality Hub managed runs — all from inside a chat session

The MCP server runs entirely on your local machine. No project files, test code, or credentials are transmitted to Provar servers.

Before you can use Provar MCP, ensure the following are in place:

Provar MCP requires an active Provar Automation license on the machine where the server runs. Validation is automatic:

1. The server reads ~/Provar/.licenses/*.properties — the same files written by the Provar Automation IDE — and checks that a license is activated and was last verified online within 48 hours.
2. Successful validations are cached for 2 hours, so frequent server restarts do not cause repeated disk reads.
3. If no valid license is found, the server exits immediately with a clear error message. Open Provar Automation IDE and ensure your license is activated, then retry.

There is no separate MCP license. Your existing Provar Automation license covers MCP usage.

npm install -g @salesforce/cli
sf --version

sf plugins install @provartesting/provardx-cli
sf provar mcp start --help

Run sf provar auth login to connect your Provar account and unlock full Quality Hub API validation (170+ rules, quality scoring). Without this, the MCP server runs in local-only mode using structural rules.

sf provar auth login

This opens a browser to the Provar login page. After you authenticate, your API key is stored at ~/.provar/credentials.json and picked up automatically by the MCP server on every subsequent tool call.

For CI/CD pipelines (GitHub Actions, Jenkins, etc.) where a browser cannot open: run sf provar auth login once on your local machine, copy the api_key value from ~/.provar/credentials.json, and store it as the PROVAR_API_KEY environment variable or secret in your pipeline. The key is valid for approximately 90 days — rotate the secret when it expires by running sf provar auth login again locally.

Edit the Claude Desktop MCP configuration file:

• macOS / Linux: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "provar": {
      "command": "sf",
      "args": ["provar", "mcp", "start", "--allowed-paths", "/path/to/your/provar/project"]
    }
  }
}

Restart Claude Desktop after saving. The Provar tools will appear in the tool list automatically.

Claude Code can be configured via the claude CLI command or by editing a JSON config file. Both approaches work in the terminal, the VS Code extension, and the Claude Code Desktop app.

Via terminal (recommended):

# User-scoped — works across all your projects
claude mcp add provar -s user -- sf provar mcp start --allowed-paths /path/to/your/provar/project

# Project-scoped, shared — creates .mcp.json at project root; commit to source control
claude mcp add provar -s project -- sf provar mcp start --allowed-paths /path/to/your/provar/project

Via config file — create .mcp.json at your project root:

{
  "mcpServers": {
    "provar": {
      "command": "sf",
      "args": ["provar", "mcp", "start", "--allowed-paths", "/path/to/your/provar/project"]
    }
  }
}

sf not found? GUI environments (VS Code, Claude Code Desktop) often launch with a restricted PATH that doesn't include sf. Use npx as the command instead:

# Terminal
claude mcp add provar -s user -- npx -y @salesforce/cli provar mcp start --allowed-paths /path/to/your/provar/project

{
  "mcpServers": {
    "provar": {
      "command": "npx",
      "args": ["-y", "@salesforce/cli", "provar", "mcp", "start", "--allowed-paths", "/path/to/your/provar/project"]
    }
  }
}

Create .vscode/mcp.json in your workspace root (commit to share with your team):

{
  "servers": {
    "provar": {
      "type": "stdio",
      "command": "sf",
      "args": ["provar", "mcp", "start", "--allowed-paths", "${workspaceFolder}"]
    }
  }
}

Open the GitHub Copilot Chat panel and switch to Agent mode. The Provar tools will appear in the tool list.

sf not found? Replace "command": "sf" with "command": "npx" and prepend "-y", "@salesforce/cli" to the args array.

Add to .cursor/mcp.json in your workspace root (project-level) or ~/.cursor/mcp.json (global):

{
  "mcpServers": {
    "provar": {
      "command": "sf",
      "args": ["provar", "mcp", "start", "--allowed-paths", "/path/to/your/provar/project"]
    }
  }
}

Restart Cursor after saving. The Provar tools will appear under Settings → MCP.

sf not found? Replace "command": "sf" with "command": "npx" and prepend "-y", "@salesforce/cli" to the args array.

Important: Set --allowed-paths to the root of your Provar Automation project directory (the folder containing your .testproject file). The server will only read and write files within this boundary.

Once your AI client is configured, ask it:

"Call provardx_ping with message 'hello'"

Expected response:

{ "pong": "hello", "ts": "2026-04-07T...", "server": "provar-mcp@1.0.0" }

If this fails, see the Troubleshooting section.

Get an instant inventory of your Provar project — file counts, coverage gaps, and missing configurations.

Prompt:

"Use provar_project_inspect on my project at /workspace/MyProvarProject and tell me what you find — how many test cases are there, and which ones aren't covered by any test plan?"

What you get back:

• Total test case count, suite structure, Page Object count
• A list of test cases not referenced by any test plan (coverage gaps)
• Whether a provardx-properties.json config file exists

Score an existing test case for schema compliance and best-practice quality issues.

Prompt:

"Validate the test case at /workspace/MyProvarProject/tests/regression/LoginTest.testcase and explain any issues."

What you get back:

• validity_score (schema compliance, 0–100) and quality_score (best practices, 0–100)
• Specific rule violations with IDs, severities, and descriptions
• Actionable suggestions (e.g. "Add a missing XML declaration", "Test case ID is not a valid UUID")
• validation_source — "quality_hub" if authenticated, "local" if no API key is configured

Get more: Run sf provar auth login once to unlock Quality Hub API validation (170+ rules). Without a key the tool still returns useful results using local structural rules.

Have the AI scaffold a new Java Page Object for a Salesforce page with correct annotations and @FindBy stubs.

Prompt:

"Generate a Salesforce Page Object for the Account Detail page. Include fields for Account Name (input), Industry (select), and a Save button. Write it to /workspace/MyProvarProject/src/pageobjects/accounts/AccountDetailPage.java."

What you get back:

• A valid Java file with @SalesforcePage annotation
• @FindBy annotations for each field using sensible locator strategies
• File written to disk (use dry_run: true in the tool call to preview without writing)

Scaffold a new XML test case with a proper UUID, sequential step IDs, and a clean structure ready for Provar Automation.

Prompt:

"Generate a test case called 'Verify Account Creation' with steps for navigating to the Accounts page, clicking New, filling in Account Name, and saving. Write it to /workspace/MyProvarProject/tests/smoke/VerifyAccountCreation.testcase."

Let the AI create and validate a provardx-properties.json — the properties file that tells the Provar DX CLI how to run your tests.

Prompt:

"Generate a provardx-properties.json at /workspace/MyProvarProject/provardx-properties.json with projectPath set to /workspace/MyProvarProject and provarHome set to /Applications/Provar. Then validate it and tell me if anything is missing."

Get a single quality score for your entire project — test cases, suites, plans, connections, environments, and cross-cutting rules all evaluated together.

Prompt:

"Validate the full test project at /workspace/MyProvarProject. The project has connections named SandboxOrg and ProdOrg, and environments QA and UAT. Give me a quality report."

What you get back:

• Overall project quality score (0–100)
• Test plan coverage percentage
• Breakdown of violations by rule ID
• Per-plan quality scores

Ask the AI to run your local Provar Automation test suite and report results.

Prompt:

"Load the properties file at /workspace/MyProvarProject/provardx-properties.json, compile the project, then run the tests and tell me the results."

The AI will chain:

1. provar_automation_config_load — registers the properties file
2. provar_automation_compile — compiles Page Objects
3. provar_automation_testrun — executes the test run
4. provar_testrun_report_locate — finds the JUnit/HTML report paths

Kick off a managed test run via Provar Quality Hub and poll until it completes.

Pre-requisite: Authenticate the Salesforce CLI against your Quality Hub org first:

sf org login web -a MyQHOrg
sf provar quality-hub connect -o MyQHOrg

Prompt:

"Connect to the Quality Hub org MyQHOrg, start a test run using config file config/smoke-run.json, and poll every 30 seconds until it completes or fails."

The AI will chain:

1. provar_qualityhub_connect — connects to the org
2. provar_qualityhub_testrun — triggers the run
3. provar_qualityhub_testrun_report — polls status in a loop
4. Reports final pass/fail status and a summary of results

After a failed run, ask the AI to classify failures and identify patterns.

Prompt:

"My test run just finished. Analyse the results at /workspace/MyProvarProject/Results/ and classify any failures — tell me which are pre-existing issues and which look like new regressions."

What you get back:

• Classified failure categories (environment issue, assertion failure, locator issue, etc.)
• Identification of Page Objects involved in failures
• Suggested next steps

Turn a failed test execution directly into a Quality Hub defect, without leaving your AI chat.

Prompt:

"The test 'LoginTest' failed in the last run. Create a defect in Quality Hub for it."

• Local only. The MCP server communicates via stdio — no TCP port is opened, no network listener is started.
• Path-scoped. All file operations are restricted to the directories you specify via --allowed-paths. Path traversal (../) is blocked.
• No data exfiltration. Project files, test code, and credentials are never transmitted to Provar servers.
• Credential safety. Quality Hub and Automation tools invoke the Salesforce CLI as a subprocess. Org credentials stay in the SF CLI's own credential store and are never read or logged by the MCP server.
• Audit log. Every tool invocation is logged to stderr with a unique request ID in structured JSON format. Capture stderr to maintain an audit trail.

"No activated Provar license found" / LICENSE_NOT_FOUND Open Provar Automation IDE → Help → Manage License → ensure the license is Activated. Then restart the MCP server.

"Warning: license validated from offline cache" (on stderr) The server started successfully but the license cache is over 2 hours old. This is a warning only. If the cache exceeds 48 hours without a successful online re-validation, the next startup will fail. Restart the server while Provar Automation IDE is connected to the internet to refresh the cache.

SF_NOT_FOUND error from Quality Hub / Automation tools The sf CLI binary is not on the PATH that the MCP server sees (common with macOS and Windows GUI apps). Use npx as the command in your MCP config — it resolves @salesforce/cli from your npm cache without needing sf on PATH:

{ "command": "npx", "args": ["-y", "@salesforce/cli", "provar", "mcp", "start", "--allowed-paths", "..."] }

Alternatively, use the full path to the sf binary (e.g. /usr/local/bin/sf on macOS).

PATH_NOT_ALLOWED error The path passed to a tool is outside the --allowed-paths root. Update --allowed-paths in your client config and restart the server.

Tools not appearing in Claude Desktop After editing claude_desktop_config.json, fully quit and reopen Claude Desktop (Cmd+Q on macOS, not just close the window).

Server starts then immediately exits Check the plugin is installed: sf plugins | grep provardx. If missing: sf plugins install @provartesting/provardx-cli.

• Bug reports and feature requests (COMING SOON): github.com/ProvarTesting/provardx-cli/issues
• Provar Automation / Quality Hub support: Contact Provar Support through your usual channel or through the Provar Success Portal.