5 MCP Automation Recipes Every Developer Should Steal in 2026

TL;DR — Key Takeaways

• Playwright MCP (official Microsoft package: @playwright/mcp) lets Claude run browser tests, see failures, fix code, and retest — all without leaving your editor
• GitHub MCP has 51 tools and 28,300+ stars — Claude can read issues, write code, commit, open a PR, and request review in one natural language command
• Supabase MCP exposes 20+ tools — schema inspection, migration generation, and RLS policy updates from plain English; always use read_only=true on production
• Slack MCP lets Claude post summaries, search conversations, and notify channels — combine with GitHub MCP for automated PR digests
• Most developers run 3–5 MCP servers in parallel and chain them for end-to-end workflows; the recipes below are copy-paste ready

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest"]
    }
  }
}

Prompt pattern

"Navigate to localhost:3000/checkout, fill in the test card (4242 4242 4242 4242), submit the form, and verify the success screen appears. If any step fails, fix the component and try again."

Token cost to know about

Playwright MCP uses roughly 114,000 tokens per typical task vs ~27,000 for the Playwright CLI. In Cursor, Windsurf, or Claude Desktop (no filesystem access) MCP is your only option. In Claude Code with filesystem access, consider the CLI for large test suites to control costs.

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "your-token-here"
      }
    }
  }
}

Prompt pattern — issue to PR

"Find the open issue about the broken password reset flow. Create a branch called fix/password-reset, fix the bug, write a test, commit with a descriptive message, and open a PR targeting main. Assign it to me and add the label 'bug'."

Prompt pattern — issue triage

"Find all open issues labelled 'bug' older than 7 days. Summarize each one, and add a comment on each saying it's been flagged for triage this sprint."

Already have a GitHub MCP server running?

Test its 51 tools in the browser — no install, no sign-up.

{
  "mcpServers": {
    "supabase": {
      "command": "npx",
      "args": [
        "-y",
        "@supabase/mcp-server-supabase@latest",
        "--access-token",
        "your-access-token"
      ]
    }
  }
}

Production safety — read this before connecting

The MCP server connects with the service role key, which bypasses RLS and has admin-level access to all data. For production databases, always use read_only=true in the connection config. Test every schema change on a staging project first.

Prompt pattern — schema migration

"Add a subscription_tier column to the users table — enum with values free, pro, enterprise, defaulting to free. Update the RLS policies so users can only read their own row. Generate the migration and apply it."

Prompt pattern — security audit

"Audit the RLS policies on the orders table and flag any that allow unauthenticated reads or writes."

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "your-token" }
    },
    "slack": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-slack"],
      "env": {
        "SLACK_BOT_TOKEN": "xoxb-your-token",
        "SLACK_TEAM_ID": "T00000000"
      }
    }
  }
}

Prompt — daily shipping digest

"Pull all PRs merged in the last 24 hours from acme-corp/backend. Write a one-paragraph summary of what shipped — mention the authors by name. Post it to #engineering in Slack."

Prompt — stale PR reminder

"Find PRs open for more than 3 days with no review activity. Post a reminder to #code-review tagging the assigned reviewers."

The workflow — step by step

Single prompt that kicks it all off

"Pick up GitHub issue #142. Fix the subscription tier filter bug — write the component, add a DB index, test it with Playwright on localhost, open a PR, and notify #product in Slack when it's ready for review."

Before you chain servers in production

Test each MCP server individually first. Connect it in MCP Playground, list its tools, and run a few manual calls. Understanding what each server can and can't do prevents surprises when Claude starts chaining them autonomously.

Test your MCP servers before chaining them

Connect any MCP server, inspect its tools, and run test calls — free, no sign-up required.