diff --git a/docs/investigations/playwright-mcp-startup-investigation.md b/docs/investigations/playwright-mcp-startup-investigation.md
new file mode 100644
index 00000000000..0250fee1292
--- /dev/null
+++ b/docs/investigations/playwright-mcp-startup-investigation.md
@@ -0,0 +1,132 @@
+# Playwright MCP Startup Investigation
+
+**Date**: 2026-02-03  
+**Issue**: Playwright MCP "calling initialize: EOF" error during gateway startup  
+**Workflow**: daily-multi-device-docs-tester  
+**Run ID**: 21638629201  
+**Status**: ✅ Resolved - Cosmetic error, no functional impact
+
+## Executive Summary
+
+The Playwright MCP server reports an "EOF" error during MCP Gateway startup, but this is a **cosmetic issue** that does not affect workflow functionality. Playwright successfully starts and operates correctly when the agent actually connects.
+
+## Timeline
+
+| Time | Event | Status |
+|------|-------|--------|
+| 16:35:03 | Gateway pre-registration attempts to initialize Playwright | ❌ Failed with EOF |
+| 16:35:08 | Gateway health check | ✅ Reports Playwright running |
+| 16:35:48 | Claude agent connects to Playwright via HTTP | ✅ Connection successful |
+| 16:35:48-16:39:39 | Playwright tools used for multi-device testing | ✅ All operations succeed |
+| 16:39:39 | Workflow completes with test results | ✅ Success |
+
+## Root Cause
+
+### The Problem
+
+During startup, the MCP Gateway attempts to **eagerly initialize** all configured MCP servers to pre-register their tools. This process:
+
+1. Launches the Playwright Docker container
+2. Attempts immediate stdio handshake via MCP protocol
+3. Fails with "calling 'initialize': EOF" 
+4. Reports the error but continues gateway startup
+
+### Why It Fails
+
+The Playwright MCP container needs a brief moment to:
+- Initialize the Node.js process
+- Set up browser binaries
+- Start the MCP protocol handler
+
+**Gateway Timeout**: The initialization happens too quickly, causing an EOF when the container hasn't fully started its stdio handler.
+
+### Why It Works Anyway
+
+The MCP Gateway uses **lazy initialization** as a fallback:
+
+1. Even though pre-registration fails, the gateway **keeps the `/mcp/playwright` route active**
+2. When the agent connects later (at 16:35:48, 45 seconds after startup)
+3. Playwright is fully initialized and responds correctly
+4. All tools are available and functional
+
+## Evidence
+
+### 1. Error During Startup
+
+```log
+2026/02/03 16:35:03 [LAUNCHER] Starting MCP server: playwright
+2026/02/03 16:35:03 ❌ MCP Connection Failed:
+2026/02/03 16:35:03    Error: calling "initialize": EOF
+2026/02/03 16:35:03    ⚠️  Process started but terminated unexpectedly
+```
+
+### 2. Gateway Continues Successfully
+
+```log
+2026/02/03 16:35:03 Starting MCPG in ROUTED mode on 0.0.0.0:80
+2026/02/03 16:35:03 Routes: /mcp/<server> where <server> is one of: [safeoutputs github playwright]
+2026/02/03 16:35:03 Registered route: /mcp/playwright
+```
+
+### 3. Agent Connection Succeeds
+
+```log
+2026-02-03T16:35:48Z [DEBUG] MCP server "playwright": Successfully connected in 85ms
+2026-02-03T16:35:48Z [DEBUG] MCP server "playwright": Connection established
+```
+
+### 4. Tools Actually Work
+
+The workflow successfully:
+- Tested 10+ device configurations (mobile, tablet, desktop)
+- Navigated to multiple pages
+- Took screenshots
+- Ran accessibility checks
+- Generated comprehensive test results
+
+## Impact Assessment
+
+### User Impact
+
+**None** - The error message appears in logs but:
+- ✅ Workflows execute successfully
+- ✅ Playwright tools are fully functional
+- ✅ Results are generated correctly
+- ✅ No retry or manual intervention needed
+
+### Operational Impact
+
+**Low** - May cause confusion:
+- ⚠️ Error messages in gateway logs look alarming
+- ⚠️ Pre-registration failure might mask real issues
+- ⚠️ Monitoring alerts could trigger incorrectly
+
+## Recommendations
+
+### Short Term (Completed)
+
+1. ✅ **Document this behavior** - This investigation serves as documentation
+2. ✅ **Verify no functional impact** - Confirmed through artifact analysis
+
+### Medium Term (Proposed)
+
+1. **Improve error messaging** in MCP Gateway:
+   ```
+   ⚠️ Pre-registration failed for 'playwright' (will retry on first connection)
+   ```
+   Instead of:
+   ```
+   ❌ FAILED to launch server 'playwright'
+   ```
+
+2. **Add retry logic** with exponential backoff
+
+3. **Distinguish error types**:
+   - "Pre-registration timeout" (non-critical)
+   - "Server unavailable" (critical)
+
+## Conclusion
+
+The "Playwright MCP not properly started" error is a **false alarm**. The MCP Gateway's eager initialization strategy encounters a timing issue with Playwright's container startup, but the lazy initialization fallback ensures everything works correctly when the agent actually connects.
+
+**No code changes are required** for functionality. Improvements to error messaging and retry logic would enhance observability but are not critical.