Can I Use MCP With Gemini? What Works In The App, The CLI, And Google Cloud Right Now
Gemini support for MCP is not one single thing, it is spread across three very different products with three very different setup paths. I tested each one to figure out which actually applies to you.
Marcus Webb
July 11, 2026
1. Introduction
Yes, Gemini supports MCP, but where you access it depends heavily on which Gemini product you actually mean. The consumer chat app at gemini.google.com, the Gemini CLI command line tool, and Google Cloud's managed MCP servers are three separate surfaces with different maturity levels, and conflating them is the fastest way to end up confused about why a feature you read about does not appear where you are looking.
2. The Problem: Three Products, One Name
Most of the concrete, hands on MCP documentation from Google is written for Gemini CLI and for Google Cloud's Apigee and BigQuery integrations aimed at developers and enterprises, not for the everyday consumer chat interface. If your mental model of MCP comes from Claude's connector directory or ChatGPT's Developer Mode, both of which live inside the main chat product, Gemini's approach can feel scattered by comparison because the most complete support currently sits in the developer tooling rather than the consumer app.
3. Causes And Fixes: Every Common Gemini MCP Issue
- Looking for an MCP connectors menu inside the regular Gemini chat app and not finding one: this is expected, the most mature MCP support currently runs through Gemini CLI, not the consumer web chat
- Gemini CLI not discovering your MCP server's tools: check your settings.json or mcp_config.json configuration is correctly pointing to your server, and confirm the server name does not contain an underscore, which breaks the CLI's name parsing
- Environment variables not reaching your MCP server process: Gemini CLI sanitizes sensitive variables like API keys and tokens by default for security, you need to explicitly whitelist any variable your server actually needs in the env property of its config
- Trying to connect a local MCP server to a managed Google Cloud agent: the fully managed remote MCP servers are built for cloud hosted agents using tools like Apigee, not for pointing directly at your laptop, use Gemini CLI instead for local development work
- Tool responses returning as plain text when you expected rich content: confirm your server's response follows the MCP specification for structured content blocks, Gemini CLI does support images and binary data but only if the server formats the response correctly
4. Examples: A Working Gemini CLI MCP Setup
// Example Gemini CLI MCP server configuration
// Place inside settings.json or mcp_config.json
{
"mcpServers": {
"my-database-tool": {
"command": "node",
"args": ["./mcp-server/index.js"],
"env": {
"DATABASE_URL": "$DATABASE_URL"
}
}
}
}Once this configuration is saved, Gemini CLI fetches the tool definitions from your server on startup, validates the schemas for compatibility with the Gemini API, and registers them so the model can call them like any built in tool. From there you can reference the tool naturally in a prompt and Gemini CLI handles the confirmation and execution flow automatically.
5. Common Mistakes
- Expecting the consumer Gemini app to have the same connector ecosystem as Claude or ChatGPT today, the current center of gravity for Gemini MCP support is developer tooling, not the chat app
- Using underscores in MCP server names inside Gemini CLI configuration, which silently breaks wildcard security rules
- Assuming any environment variable your server needs will pass through automatically, Gemini CLI blocks sensitive looking variables by default
- Pointing a managed Google Cloud MCP server setup at a local development server instead of using Gemini CLI, which is the actual tool designed for local server development
6. Best Practices
If you are experimenting or building your own MCP server, start with Gemini CLI, since it has the most complete and well documented client implementation right now. If you are working inside an enterprise Google Cloud environment and want to expose BigQuery, Cloud Run, or other Google services to an agent, use Google's fully managed remote MCP servers instead of building your own from scratch, since Google is actively expanding coverage across more services. For the consumer Gemini app specifically, check Google's own release notes before assuming a feature exists, since this is the surface changing fastest and least documented in third party guides.
7. FAQ
- Can I connect a custom MCP server to the regular Gemini app the way I can with Claude: not with the same maturity level as Claude's connector directory today, Gemini's strongest MCP support currently lives in Gemini CLI and Google Cloud tooling
- Is Gemini CLI free to use: yes, it is open source and free, though the underlying Gemini API usage may have its own costs depending on your usage volume
- Do I need a Google Cloud account to use MCP with Gemini CLI: no, for local MCP servers you only need Gemini CLI itself, a Google Cloud account is only needed for Google's fully managed remote MCP servers
- Will an MCP server built for Claude work with Gemini CLI: often yes if it follows the standard MCP specification, but always test it, since schema handling can differ slightly between clients
8. Conclusion
Gemini does support MCP, and the developer facing tooling around it is genuinely solid, but the experience is currently split across Gemini CLI and Google Cloud rather than centralized in the consumer chat app the way it is with Claude or ChatGPT. If you came looking for a connectors button inside gemini.google.com, that is not where the real capability sits today, Gemini CLI is.