Gemini CLI v0.40 Ships Offline Ripgrep, MCP Resource Tools, and a Rebuilt Memory System

Gemini CLI v0.40.0 shipped as the new stable release on April 28. This follows v0.39, which landed on April 23 with the memory inbox and skill patching system.

v0.40 is a different kind of release. Where recent versions focused on what the agent learns and remembers, this one focuses on what it can reach. Bundled ripgrep for offline code search, native MCP resource tools, and a rebuilt memory architecture are the headline changes.

Offline Ripgrep Support

Gemini CLI now bundles the ripgrep binary inside its Single Executable Application. That means code search works without a separate ripgrep installation and without network access.

This matters for two different developer situations. First, offline or air-gapped environments, which are common in enterprise settings, where outbound internet access from developer workstations is restricted. Second, any environment where installing additional system binaries requires an approval process or admin access. Bundling ripgrep removes that dependency entirely.

The practical result is that grep-based code search now works out of the box across any platform where Gemini CLI runs.

MCP Resource Tools

v0.40 adds two new core tools: list MCP resources and read MCP resources.

These tools let the agent interact with the resources exposed by any connected MCP server, not just its tools. The distinction matters because MCP servers can expose two kinds of things: tools (callable functions) and resources (readable data like files, schemas, database contents, or API responses). Until now, Gemini CLI could only invoke MCP tools. With v0.40, it can also inspect and read MCP resources directly.

The practical use case is agents that need to reason about the state of connected systems before deciding what to do, not just call their APIs. For teams running MCP servers for databases, internal APIs, or custom data sources, this significantly expands what the agent can introspect without requiring a custom tool for every data type.

Memory System Rebuilt

The memory management system has been rewritten. The previous MemoryManagerAgent is gone, replaced with a prompt-driven four-tier system.

The tiers in the new system separate memory by scope and persistence: session-level (current conversation only), project-level (shared across the project), user-level (follows the user across projects), and global (persists across all contexts). Each tier uses prompt-driven editing rather than a separate agent managing memory as a background process.

The practical effect is more predictable memory behavior. The previous system’s separate agent could conflict with session state or apply memory changes at unexpected times. The prompt-driven approach integrates memory management into the normal flow of the agent’s reasoning, making it easier to understand what the agent knows and why it is acting on that knowledge.

The v0.39 skill extraction and patching system is still in place. The v0.40 rebuild sits underneath it, providing the storage and retrieval layer that skill patterns write into.

Streamlined Local Gemma Setup

The gemini gemma command now handles local model setup more cleanly. The previous flow required several manual steps to configure a local Gemma model for inference. The updated command reduces this to a single streamlined path.

This is aimed at developers who want to run Gemini CLI against a local model rather than routing everything through Google’s API, which is increasingly common for privacy-sensitive workflows or for reducing API costs on high-volume tasks.

Other Changes

A few additional things in v0.40:

• /new is now an alias for /clear, giving a slightly more intuitive command for starting a fresh session
• GitHub colorblind themes are available in the terminal UI
• OSC 777 terminal notification support for session events
• Vertex AI request routing settings for enterprise Google Cloud deployments
• Protection against workspace .env files overriding sensitive environment variables

The last item is a security fix. A malicious or misconfigured workspace .env file could previously override environment variables that control authentication and API routing. v0.40 blocks that override for sensitive variables, which is the kind of change that matters in shared repository or multi-developer environments.

v0.40.0 is available via npm update -g @google/gemini-cli or through the standard install path at geminicli.com.

---

Sources: Gemini CLI Changelog, GitHub Release v0.40.0