How to fix MCP error "spawn npx ENOENT / command not found"?

The AI host application (Claude Desktop, Cursor, etc.) was launched in a graphical desktop environment that does not inherit your terminal's user $PATH, or Node.js is not installed globally. Quick Fix: Replace 'command': 'npx' with the absolute path to your npx binary (e.g. C:/Program Files/nodejs/npx.cmd on Windows or /usr/local/bin/npx on macOS).

How to fix MCP error "JSON-RPC error -32603: Internal error (STDOUT Pollution)"?

The MCP server wrote non-JSON debug strings to standard output (STDOUT) instead of standard error (STDERR). In stdio transport, STDOUT is strictly reserved for JSON-RPC messages; any `console.log()` breaks JSON parsing. Quick Fix: Switch all logging inside your custom MCP server from `console.log()` to `console.error()`. For Python servers, write logs exclusively to `sys.stderr`.

How to fix MCP error "JSON-RPC error -32602: Invalid params"?

The AI model passed arguments with incorrect data types (e.g. string instead of integer or array), or omitted a required parameter defined in inputSchema. Quick Fix: Inspect the tool's inputSchema with tools/list and ensure all required fields match their declared types.

How to fix MCP error "JSON-RPC error -32700: Parse error / Invalid Windows backslashes"?

Single backslashes used in Windows file paths (e.g. 'C:\Projects') violate JSON string escape rules. Quick Fix: Replace single backslashes with forward slashes (e.g. 'C:/Projects') or double-escaped backslashes ('C:\\Projects').

How to fix MCP error "SSE connection closed prematurely / EOF"?

Reverse proxies (Cloudflare, Nginx, ALB) or corporate firewalls terminate idle HTTP streaming connections after 60–100 seconds when keep-alive heartbeats are missing. Quick Fix: Implement a 15-second ping/heartbeat comment (`: keep-alive\n\n`) on the SSE endpoint and disable proxy buffering (`X-Accel-Buffering: no`).

Developer Troubleshooting EngineInstant OS Fixes

MCP Error Decoder

Search exact error strings and stack traces from Claude Desktop, Cursor, and OpenAI Codex CLI. Inspect root causes and copy verified terminal fixes.

Scan Config in Linter

Category:

Performance & DatabaseMedium

Tool call timed out after 60000ms

Symptom

Long-running database queries, video processing, or web scraping tool calls abort with a timeout error.

Root Cause

Default client timeouts limit tool execution duration to 60 seconds before concluding the stdio process is unresponsive.

Remediation & Quick Fix

Split heavy queries into paginated chunks or implement asynchronous polling jobs where the tool returns a job ID immediately.

OS Diagnostic Commands:

mac

# Check query plan latency before running in MCP:
psql $DATABASE_URL -c "EXPLAIN ANALYZE SELECT ...;"

windows

# Check query plan latency:
psql.exe $env:DATABASE_URL -c "EXPLAIN ANALYZE SELECT ...;"

linux

# Check query plan latency:
psql $DATABASE_URL -c "EXPLAIN ANALYZE SELECT ...;"

Verified Configuration Example:

// In your MCP server tool handler:
// Return job token immediately rather than hanging for 120 seconds
return {
  content: [{ type: "text", text: "Job submitted with ID: task_9821. Call check_status to poll." }]
};