Windsurf MCP Server for Bookmarks (Cascade Setup, 2026)

Windsurf is an AI code editor from Codeium, built around its Cascade AI agent. Cascade supports MCP (Model Context Protocol), meaning it can connect to external tools and data sources during coding sessions — including your ContextBolt bookmarks.

Once connected, Cascade can search your saved web pages and your saves from Twitter/X, Reddit and LinkedIn without you leaving the editor. This is particularly useful during Cascade’s agentic coding flows, where it can autonomously pull in relevant bookmarks as context for the code it’s generating.

Why connect bookmarks to Windsurf

Cascade is designed to handle multi-step coding tasks autonomously. Adding your bookmarks to its toolkit means it can reference your curated knowledge when making decisions.

Say you’re building an authentication system. You’ve saved tweets about JWT best practices, a Reddit thread about OAuth pitfalls, and a LinkedIn post about session management. With the ContextBolt integration, Cascade can search those saves as part of its workflow, grounding its code generation in material you’ve already vetted.

This is different from web search. Your bookmarks are content you specifically chose to save because you found it valuable. That’s what makes them a second brain. That signal matters — it means the results are pre-filtered by your own judgment.

The full Windsurf MCP config block

Use https://api.contextbolt.app/mcp and complete the browser sign-in if your client supports remote MCP with OAuth. Enter your Pro account email and email code, then allow the connection. For a client that cannot complete that flow, use the personal-URL compatibility configuration shown here. Copy the full private URL from the extension under Settings, MCP Integration, Older clients and replace the example URL with it.

Drop this into ~/.codeium/windsurf/mcp_config.json on macOS or Linux, or %USERPROFILE%\.codeium\windsurf\mcp_config.json on Windows. Replace the complete https://api.contextbolt.app/mcp/YOUR_TOKEN example with your private URL from Settings → MCP Integration. Newer versions keep it under Older clients. If the file does not exist, create it.

{
  "mcpServers": {
    "bookmarks": {
      "serverUrl": "https://api.contextbolt.app/mcp/YOUR_TOKEN"
    }
  }
}

The field is serverUrl, not url. We tested both during the rollout: copying a Cursor-style block ("url") into Windsurf produces no error and no tools, which means the connection looks fine in the file but Cascade never sees the bookmarks server. The Cascade MCP documentation covers the field name and the wider config schema.

Tips for Windsurf workflows

Let Cascade discover relevant saves: During agentic coding sessions, Cascade can decide on its own to search your bookmarks when the task seems related to content you might have saved. The more descriptive your task prompts, the better it can match.

Start sessions with context: Before a big task, ask Cascade “Check my bookmarks for anything about [topic]” to front-load your saved knowledge into the conversation.

Use for architecture decisions: When Cascade suggests an approach, ask it to search your bookmarks for alternative patterns. Your saves might contain a better solution than the default.