Claude Code CLI Hangs on Startup with Ollama (Anthropic Messages API)

[virtualassistanterion]

Summary

Claude Code CLI (claude --model qwen2.5-coder:14b -p) hangs indefinitely during startup when configured to use a local Ollama instance via Anthropic Messages API protocol.

Environment

• Claude Code Version: 2.1.39
• OS: Linux (Debian/Ubuntu)
• Node.js: v24.13.1
• Ollama Version: v0.15.6
• Model: qwen2.5-coder:14b (14.8B, Q4_K_M)

Configuration

export ANTHROPIC_AUTH_TOKEN=ollama
export ANTHROPIC_BASE_URL=http://localhost:11434
claude --model qwen2.5-coder:14b "Hello" -p

Expected Behavior

• Claude Code should initialize the session and respond to the prompt within 5-10 seconds
• Output should be printed and the process should exit (due to -p flag)

Actual Behavior

• Process hangs indefinitely after initialization
• Debug logs show it's stuck loading MCP servers from https://api.anthropic.com/v1/mcp_servers?limit=1000
• No error messages or timeout errors
• Process must be manually killed (SIGKILL)
• Works fine with default Anthropic API key (ANTHROPIC_API_KEY)

Debug Output

2026-02-13T00:32:19.854Z [DEBUG] [STARTUP] Loading MCP configs...
2026-02-13T00:32:19.857Z [DEBUG] Found 0 plugins (0 enabled, 0 disabled)
2026-02-13T00:32:19.857Z [DEBUG] [claudeai-mcp] Checking gate (cached)...
2026-02-13T00:32:19.857Z [DEBUG] [claudeai-mcp] Gate returned: true
2026-02-13T00:32:19.857Z [DEBUG] [claudeai-mcp] Fetching from https://api.anthropic.com/v1/mcp_servers?limit=1000
(HANGS HERE - never completes)

Workaround

Using sessions_spawn with local model or direct Ollama invocation works perfectly:

ollama run qwen2.5-coder:14b "Hello"  # Works fine

Root Cause Hypothesis

Claude Code CLI appears to be attempting to fetch MCP server configurations from the Anthropic cloud endpoint (api.anthropic.com) even when using a local Ollama backend. This causes a hang because the connection to the cloud endpoint is never resolved with a timeout.

Steps to Reproduce

1. Install Ollama (https://ollama.ai)
2. Pull a model: ollama pull qwen2.5-coder:14b
3. Set environment variables:

   export ANTHROPIC_AUTH_TOKEN=ollama
   export ANTHROPIC_BASE_URL=http://localhost:11434

4. Run: claude --model qwen2.5-coder:14b "Hello" -p
5. Observe: Process hangs for >60 seconds and must be killed

Suggested Fixes

1. Add startup timeout: Implement a configurable timeout (e.g., 10s) for MCP server fetching
2. Skip cloud MCP calls for local backends: Detect when ANTHROPIC_BASE_URL is localhost and skip cloud MCP discovery
3. Fallback mechanism: If MCP fetch fails/times out, log warning and continue with base tools only
4. Option flag: Add --skip-mcp or --mcp-disabled flag for users who don't need MCP servers

[github-actions]

Found 2 possible duplicate issues:

1. [BUG] CLI hangs indefinitely on simple queries (macOS M1, v2.1.37) #24481
2. [CRITICAL] MCP Server Causes 16+ Hour Hang - No Timeout or Stuck Detection #15945

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code