Windsurf Integration

Give Windsurf's Cascade AI persistent memory across every coding session. Dakera's MCP server connects directly to Windsurf — Cascade remembers project architecture, coding decisions, debugging context, and team patterns so you never have to re-explain your codebase.

Quick Start

Get persistent memory working in Windsurf in under 2 minutes:

docker run -d --name dakera \
  -p 3300:3300 \
  -e DAKERA_API_KEY=dk-your-key-here \
  -v dakera-data:/data \
  ghcr.io/dakera-ai/dakera:latest

{
  "mcpServers": {
    "dakera": {
      "command": "docker",
      "args": ["run", "--rm", "-i",
        "--network", "host",
        "ghcr.io/dakera-ai/dakera-mcp:latest"],
      "env": {
        "DAKERA_URL": "http://localhost:3300",
        "DAKERA_API_KEY": "dk-your-key-here",
        "DAKERA_AGENT_ID": "windsurf-dev",
        "DAKERA_MCP_PROFILE": "core"
      }
    }
  }
}

What Windsurf Cascade Gets

With Dakera connected, Cascade has access to 14 core MCP tools (86+ available via profiles) organized into categories:

Windsurf MCP Config File Location

Unlike Cursor (which uses a per-project .cursor/mcp.json), Windsurf uses a single global MCP config file. This means memory is available across all your Windsurf projects without per-project setup:

Use Cases for Windsurf + Cascade

Project Architecture Memory

Store architectural decisions, patterns, and conventions once. When Cascade starts a new session, it recalls the project's design philosophy without you re-explaining the entire codebase. Cascade's deep code understanding combined with Dakera's persistent memory means your AI never starts fresh.

// Cascade will automatically use these tools:
// dakera_store: "This project uses hexagonal architecture with ports and adapters"
// dakera_store: "Database migrations use sqlx with versioned SQL files"
// dakera_recall: "What's our authentication approach?"

Cross-Session Debugging Context

When debugging a complex issue over multiple sessions, Dakera preserves the entire investigation trail — what you tried, what failed, what clues you found, and which hypotheses are still open. Cascade can pick up exactly where it left off, even days later.

Multi-Project Knowledge Base

Since Windsurf's MCP config is global, you can use namespaces to maintain separate memory spaces for each project while sharing a single Dakera instance. Switch projects in Windsurf and Cascade automatically draws from the correct namespace.

{
  "mcpServers": {
    "dakera-project-a": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "--network", "host",
        "ghcr.io/dakera-ai/dakera-mcp:latest"],
      "env": {
        "DAKERA_URL": "http://localhost:3300",
        "DAKERA_API_KEY": "dk-your-key-here",
        "DAKERA_AGENT_ID": "windsurf-dev",
        "DAKERA_NAMESPACE": "project-a",
        "DAKERA_MCP_PROFILE": "core"
      }
    }
  }
}

Team Knowledge Sharing

Multiple developers can connect to the same Dakera instance with different agent IDs. Cross-agent recall lets Cascade access shared team knowledge — architecture decisions made by a colleague are available in your session — while maintaining individual context separation.

Code Pattern Library

Store reusable patterns with importance scoring. Frequently accessed patterns score higher and surface more readily in future recalls. Cascade can suggest your team's preferred patterns automatically rather than generating generic boilerplate.

Configuration Options

Profile Selection (DAKERA_MCP_PROFILE)

The DAKERA_MCP_PROFILE environment variable controls which tools Cascade sees. Profiles balance capability vs. context window cost:

# Set in your ~/.codeium/windsurf/mcp_config.json env block:
"DAKERA_MCP_PROFILE": "power"    # Options: core, power, admin, all

Verifying the Connection

After restarting Windsurf, open Cascade and ask it to list available memory tools. You should see the Dakera MCP tools in the response:

// Ask Cascade:
"What memory tools do you have available?"

// Cascade should list:
// - dakera_store
// - dakera_recall
// - dakera_search
// - dakera_session_start
// - ...and 10 more core tools

If Cascade doesn't show the tools, check that the Dakera server is running (docker ps) and that the config file path is correct for your platform.

Next Steps

• Full MCP Server documentation — all 14 core tools (86+ available via profiles) explained
• Quick Start guide — detailed setup walkthrough
• Memory Patterns — common patterns for coding workflows
• Cursor integration — same memory, per-project config

Frequently Asked Questions

Further reading: MCP Protocol Explained · Dakera MCP: Real-World Use Cases · Cursor Integration