Cline CLI v3.0.69 raises MCP startup timeout from 3 to 10 seconds, fixing silently dropped Windows servers

Cline CLI v3.0.69 raises the default MCP server startup timeout from 3 seconds to 10 seconds. According to the release notes, MCP servers launched with npx or uvx on Windows often exceeded the old limit and were silently dropped, so their tools never appeared in the agent.

The release was published by the project's github-actions account on 07 Oct at 06:12 (tag cli-v3.0.69, commit ef9430f). It is a maintenance release with no new models, pricing, or benchmarks.

What changed

MCP startup

• The startup limit for MCP servers rises from 3 seconds to 10 seconds.
• Servers launched through npx or uvx on Windows now load. Per the release notes, they previously took longer than 3 seconds and were dropped without any error.

Provider and model handling

• Claude requests sent through a custom Anthropic base URL no longer fail with a 400 error. The notes name Azure AI Foundry and corporate gateways as affected setups.
• Models that accept only certain reasoning levels no longer reject requests. The notes cite Kimi K3 as an example. The CLI now picks the closest reasoning level the model supports.
• OpenAI-compatible providers no longer print a providerOptions key 'openai-compatible' deprecation warning on every response.

Agent behavior

• If a model's response ends without a recognized finish reason, the agent now asks the model to continue once. Previously it treated the response as complete.

CLI commands

• cline config --json prints JSON again. It had been opening the interactive view, which failed outside a terminal.
• cline config --json mcp previously printed plain text. The notes list it alongside the --json fix.
• -y/--yolo is now listed in cline --help, with a warning to use it only in sandboxed environments.

The full diff is available as the compare view cli-v3.0.68...cli-v3.0.69 on GitHub.

What this means

The MCP timeout fix matters most. A 3-second limit is tight for package runners like npx and uvx, which may resolve or download dependencies on first launch. Dropping a server without an error is the worst failure mode, because users see missing tools and have no signal about the cause. Raising the limit to 10 seconds addresses the symptom. The release notes do not mention a visible warning when a server still times out, so slow servers could still disappear without explanation.

The Anthropic base URL and reasoning-level fixes point to how Cline is used in practice. Teams route Claude through Azure AI Foundry or internal gateways, and they mix models with different reasoning-parameter rules. Clamping to the nearest supported reasoning level is a pragmatic way to handle that variation as new reasoning models such as Kimi K3 appear. It also means a requested level may not be the one actually applied, so users who need exact settings should check what the model supports.

The cline config --json fix helps anyone scripting the CLI or running it in CI, where an interactive view cannot work. Surfacing --yolo in --help with a sandbox warning makes the auto-approve option easier to find and states its risk up front.

No pricing, context window, or benchmark data applies to this release.