Playwright MCP with Ollama (via ollama-mcp-bridge)

Install + handshake last tested 2026-10-22 · Local-model tests: Not yet tested

Client: Ollama + jonigl/ollama-mcp-bridge. Client versions and features come from vendor docs, checked 2026-10-22. We haven't run this setup end-to-end yet.

Prerequisites

• Ollama installed, with a tool-capable model pulled (e.g. a Qwen3 model)
• Node.js 18+ for npx
• ollama-mcp-bridge (see its README for install and current flags)

Steps

1. Install the bridge per its README.
2. Put the mcpServers config below in the bridge's config file.
3. Limit Playwright to the suggested tools with the bridge's tool include/exclude filtering (option names per the bridge README).
4. Point your chat client at the bridge instead of Ollama's port.

Config

{
  "mcpServers": {
    "playwright-mcp-microsoft": {
      "command": "npx",
      "args": [
        "-y",
        "@playwright/mcp"
      ]
    }
  }
}

Suggested tool allowlist

Picked from the real tool names in our test; whether a small model succeeds with them is not yet tested.

• Playwright MCP (Microsoft) (25 tools listed): browser_navigate, browser_snapshot, browser_click, browser_type

Measured context cost

• Playwright MCP (Microsoft): ~3,688 tokens (45% of 8k, 11% of 32k) for all 25 tools. Small-model friendly: medium.

Method: tiktoken cl100k_base over compact JSON of name+description+inputSchema (approximate). Cost for the allowlist subset alone: not yet measured.

Success rates on small models

• Playwright MCP (Microsoft): Not yet tested

These were our own task runs, with all tools exposed. The setup on this page (client, model, allowlist) hasn't been run end-to-end.

Known failure modes (community-reported, not measured by us)

• Too many tools: the model picks the wrong one. Fix: use the allowlist above.
• Gemma thinking mode suppresses tool calls. Fix: turn thinking off.
• Context overflow from large tool schemas. Fix: fewer tools, or a bigger context window.

FAQ

Does Playwright MCP (Microsoft) work with local models?

Not yet tested. We have only run our standard install and handshake tests (last tested 2026-10-22); tool-call success with local models has not been measured.

How many tools does it expose?

Playwright MCP (Microsoft): 25 tools in our tools/list test.

Why limit the tools?

Community reports say small models choose the wrong tool or emit malformed arguments when shown many tools. We haven't measured this yet.

Are there affiliate links on this page?

No.