MCP Tools Reference

Specwright uses several MCP servers depending on your interface and inputs.

MCP Servers

Server	Package	Required?	Used by
`playwright-test`	`@playwright/mcp`	Always	`@playwright-test-planner` agent (browser automation)
`markitdown`	`markitdown-mcp-npx`	When using `filePath` input	`@process-manager` agent (PDF/DOCX conversion)
`atlassian`	Atlassian-hosted MCP	When using `inputs.jira.url`	`@process-manager` agent (Jira ticket fetching)
`specwright`	`@specwright/mcp`	Claude Desktop only	Pipeline control via natural language

Playwright MCP Tools

These tools are used internally by the @playwright-test-planner agent. You do not call them directly. They are listed here for reference when reading agent logs or debugging Phase 4 failures.

Tool	Description
`browser_navigate`	Navigate to a URL
`browser_snapshot`	Capture accessibility tree snapshot of page or element
`browser_click`	Click an element by ref or selector
`browser_type`	Type text into a focused element
`browser_select_option`	Select a dropdown option
`browser_evaluate`	Run JavaScript in browser context
`browser_close`	Close the browser

The planner agent uses browser_snapshot extensively — it produces the accessibility tree that the agent reads to discover selectors, without requiring screenshots. This is faster and more token-efficient than image-based exploration.

Specwright MCP Tools

Claude Desktop only. @specwright/mcp is not required for the Plugin + CLI or Desktop App interfaces. Only install it if you use Claude Desktop to run the pipeline.

These tools give Claude Desktop the ability to drive the full pipeline through natural language without a terminal.

Tool	Description	When used
`start_pipeline`	Start the full 10-phase pipeline	Natural language trigger
`approve_plan`	Approve the Phase 6 test plan	After reviewing the generated plan
`reject_plan`	Reject Phase 6 plan with a reason	To refine the plan before generation
`get_pipeline_status`	Check which phase is currently active	Progress check mid-run
`get_generated_files`	Read generated `.feature` and `steps.js`	After Phase 7 completes

Each tool spawns a subagent in the client project. The MCP server itself is stateless — get_pipeline_status reads from the filesystem, not from an in-memory process.

Markitdown MCP

Used by @process-manager in Phase 3 to convert local files (PDF, DOCX, PPTX) into plain text before scenario extraction.

Tool	Description
`convert`	Convert a local file to markdown text

Configured in .mcp.json:

"markitdown": {
  "command": "npx",
  "args": ["markitdown-mcp-npx"]
}

Atlassian MCP

Used by @process-manager in Phase 3 to fetch Jira ticket content (description, acceptance criteria, attachments) when inputs.jira.url is set.

Tool	Description
`fetch`	Fetch a Jira issue by URL or ticket ID
`search`	Search Jira issues using JQL

Configured as a streamable-http server pointing to Anthropic's hosted Atlassian MCP:

"atlassian": {
  "type": "streamable-http",
  "url": "https://mcp.atlassian.com/v1/mcp"
}

Authentication uses your Atlassian OAuth token or JIRA_API_TOKEN env var. The Atlassian MCP is only included in the pipeline when inputs.jira.url is set — it is not loaded for non-Jira runs.

Configuration

Project `.mcp.json`

Add this file to your project root to configure the Playwright MCP server. The plugin init command writes this automatically — only edit it manually if you need to change defaults.

{
  "mcpServers": {
    "playwright-test": {
      "command": "npx",
      "args": ["@playwright/mcp@latest", "--output-dir", ".playwright-mcp"]
    }
  }
}

The --output-dir .playwright-mcp flag keeps MCP screenshots out of your project root. The .playwright-mcp/ directory should be added to .gitignore.

Claude Desktop Config

To use the Specwright MCP server from Claude Desktop, add this to ~/.claude/claude_desktop_config.json. Set PROJECT_PATH to the absolute path of the project you want to automate.

{
  "mcpServers": {
    "specwright": {
      "command": "npx",
      "args": ["-y", "@specwright/mcp"],
      "env": { "PROJECT_PATH": "/path/to/your-project" }
    }
  }
}

Restart Claude Desktop after editing this file. You can verify connectivity by asking: "What is the current pipeline status for my project?"

← PreviousAgents Next →Shared Steps