Code assistance, structuring, memory, standby … There are many options to boost the capacity of the two online command tools.
Claude Code and Gemini Cli have vast powers. The first, the anthropic online command tool, allows developers to reshape, document and debug the code directly from the terminal. Gemini Cle is the online command line interface (CLI) of Gemini. Available in open source and free, this code assistant can for example analyze code bases, with a context of up to 1 million tokens.
But both reveal their true potential more when associated with MCP (Model Context Protocol) servers. Claude Desktop or Gemini Cli then plays the role of MCP customers. They collect and emit queries. MCP servers allow them to activate an active agent, aware of his environment and capable of performing complex tasks. Thanks to him, Claude Code and Gemini CLI can access updated information in real time, have a long -term context and perform tasks in the digital world.
Compatible MCP servers
In order to test the different MCP servers and elect the best of them on Google Claude Code or Gemini Cli, we use Gemini Cli. This is allowed because MCP servers are generally compatible with the two MCP customers. Note that the installation of MCP servers is also close to each of them. The MCPServers section is identical. The Claude Code configuration file is called Claude_desktop_Config.json. That of Gemini Cli is called Settings.json and has additional parameters.
In our case, before testing the MCP servers, we install Gemini Cli. After downloading Node.js, we go through the Windows terminal. We tap “NPM Installer -g @Google/Gemini -Cli” in the command line to set up Gemini CLI in a global way. Then, we launch Gemini Cli with the command: “Gemini”. We then connect to our Google account.
Our criteria for selection are mainly ease of installation, usefulness for developers and reliability.
1. Filesystem
The MCP Filesystem server allows you to read, write and handle files on the computer. Easy to install, it is practical for programming. Instead of constantly copying the code between the terminal and the publisher, he simply lets the agent act. What does not spoil anything is that it allows you to work in a controlled environment. Agent Gemini will not be able to access files outside the quoted root directory. The actions journalization also offers complete traceability of the actions carried out by the agent. In terms of limits of this server, note limited access to configured files and the sometimes request for appropriate permissions.
In order to configure this server, as for others, we go to the Settings.json file (on Gemini CLI). In our case, it is on Windows on the way C: \ Users \ Admin. We write this:
{
"selectedAuthType": "oauth-personal",
"theme": "Default",
"preferredEditor": "vscode",
"mcpServers": {
"filesystem": {
"command": "npx",
"args": (
"-y",
"@modelcontextprotocol/server-filesystem",
"C:\Users\admin"
)
}
}
}
Once this file has been saved, we write the ” /MCP” command in Gemini Cli. Objective: check that the MCP Filesystem server is well installed. As is the case, we ask him for different directives in natural language. For example, to understand the contents of the Settings.json file understandable.
Gemini CLI responds quickly. It sheds light on the terms of this file. This is very useful for better understanding the structure of a file.
2. Sequential Thinking
The interest of the MCP Sequential Thinkig server is to break a complex task into simpler subsets. To do this, the server can send the agent a structured list of subtaches. The latter ends up with a direction to follow. It is up to the user to validate the plan obtained. The installation on Gemini Cli is also very simple. We mention in the Settings.json file:
{
"mcpServers": {
"sequential-thinking": {
"command": "node",
"args": (
"sequential-thinking.js"
)
}
},
"selectedAuthType": "oauth-personal"
}
Once this is done, we ask the server to look at an arduous question in one of our tests:
We see in this extract that Gemini Cli breaks down the problem into sequences. He assesses the subsidies according to their criticality. It also offers practical solutions with different hypotheses. A way to have an overview of a project.
3. GitHub
The MCP Github server gives the ability to read, analyze and modify code projects hosted on the platform. These actions are carried out in an organized and secure manner via the Github API. There are many advantages. This allows in particular to have a quick overview of the different projects. It is also possible to better understand the possible concerns of code. The agent can analyze configuration files, scripts or critical input points. It can also, for example, propose code changes independently. A functionality to be used with care.
To install this server, we need a Github token. Once obtained, we are going to the Settings.json on Gemini Cli and Tapons file:
{
"mcpServers": {
"github": {
"command": "npx",
"args": ("-y", "@modelcontextprotocol/server-github"),
"env": {
"GITHUB_PERSONAL_ACCESS_TOKEN": "VOTRE_TOKEN_GITHUB"
}
}
}
}
After this stage, we simply ask the server during one of our tests information on our benchmarks:
We see in this extract that multiple structured information is displayed. A way of quickly familiarizing yourself with him. This is very interesting in the context of project management.
4. Fetch
The MCP Fetch server allows you to recover and analyze web content. It acts as a secure bridge between the agent and the outside world. It allows you to run web requests to recover the content of a web page, a API or any other resource accessible via an URL. The fact of separating this functionality into a dedicated server allows you to create a sandbox, or sandbox. This prevents a virtually corrupted agent from having direct and uncontrolled access to the network or the Internet. With this MCP server, the agent can also work faster, without having to worry about the complexity of network requests. On the limit side, note that Fetch can be blocked by certain sites.
The installation is easy, in the Settings.json on Gemini Cli file:
{
"mcpServers": {
"fetch": {
"command": "npx",
"args": ("-y", "@modelcontextprotocol/server-fetch")
}
}
}
During one of our tests, we ask him for news of the day on a particular subject.
We see that the agent went to seek various information from reliable sources. It offers a synthetic and structured summary. For example, the MCP server is practical for the day before.
5. MCP Memory
The MCP Memory server gives a persistent memory to remember the conversations. For this, he stores, collects and manages contextual information. The configuration is simple on Gemini Cli. The main commands are `Add` (add), which for example serves to remember a user preference or a technical specification, such as an API key. The `Search` (search) command allows the agent to “consult his notes”. The one named `delete` (delete) is used to erase information that has become obsolete. Note that in the event of many data to be managed, this server may have slowdowns.
To install the MCP Memory in the Settings.json on Gemini Cli file, type:
{
"mcpServers": {
"memory": {
"command": "npx",
"args": ("-y", "@modelcontextprotocol/server-memory")
}
}
}
We ask him for adding information in memory.
The request was made with the “ADD” command. Note that the parameter -Key designates the unique and permanent identifier of the stored information. It allows you to find information, update or delete it.
