The Model Context Protocol (MCP) is an open standard that lets AI assistants use external tools and data. An MCP server is a small program that exposes something, such as your files, a Git repository, GitHub or a monitoring system, as tools an AI agent can call. Claude Code, Codex CLI, many editors and desktop chat apps all speak MCP, so a server you set up once works across them.
Below are ten MCP servers that are especially useful for developers and Linux administrators. Each one is maintained by the protocol’s maintainers or by the vendor of the product it connects to. The commands use Claude Code’s claude mcp add; the same command and arguments work in any other MCP client’s configuration file.
Our top picks
- Start here#3 GitHubGitHub's hosted server covers issues, pull requests and Actions with nothing to install.See the review →
- For local repos#2 GitRead history, diffs and blame from a local repository.See the review →
- For web apps#5 PlaywrightDrives a real browser to test changes and reproduce front-end bugs.See the review →
- For incidents#9 GrafanaQueries dashboards, metrics and logs during an outage.See the review →
Before You Start
- Most servers run with npx (needs Node.js) or uvx (needs uv) and start on demand. You don’t run them as services.
- An MCP server can do whatever its tools allow, with your user’s permissions. Install servers only from sources you trust, give tokens the smallest scope possible, and keep write access off for production systems.
- In Codex CLI, add servers under [mcp_servers.name] in ~/.codex/config.toml with the same command and arguments.
1Filesystem
The reference filesystem server lets the agent read, search and write files, but only inside the directories you list. Project page: github.com/modelcontextprotocol/servers.
claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem /home/user/projects
Best for
giving desktop chat apps controlled access to a project folder.
Keep in mind
terminal agents like Claude Code can already read your working directory, so you mostly need this for chat apps or extra folders.
2Git
Reads history, diffs, branches and blame from a local repository, and can stage and commit. Project page: github.com/modelcontextprotocol/servers.
claude mcp add git -- uvx mcp-server-git --repository /home/user/projects/app
Best for
asking questions like “what changed in the deploy scripts last week?”
Keep in mind
commit tools write to the repository; review them before approving.
3GitHub
GitHub’s official server works with issues, pull requests, Actions runs, code search and repository files. GitHub hosts a remote version, so there’s nothing to install; you only need a personal access token. Project page: github.com/github/github-mcp-server.
claude mcp add --transport http github https://api.githubcopilot.com/mcp/ --header "Authorization: Bearer YOUR_GITHUB_PAT"
Best for
triaging issues, reviewing pull requests and checking why a workflow failed.
Keep in mind
use a fine-grained token limited to the repositories and permissions you need.
4Fetch
Downloads a web page and converts it to Markdown that the model can read. Project page: github.com/modelcontextprotocol/servers.
claude mcp add fetch -- uvx mcp-server-fetch
Best for
reading release notes, changelogs and documentation pages during a task.
Keep in mind
it fetches any URL the model asks for, so treat page content as untrusted input.
5Playwright
Microsoft’s Playwright server drives a real browser: it opens pages, clicks, fills forms and reads the page structure. Project page: github.com/microsoft/playwright-mcp.
claude mcp add playwright -- npx @playwright/mcp@latest
Best for
testing web apps after a change and reproducing front-end bugs.
Keep in mind
the first run downloads a browser; on a headless server, run it in headless mode as described in the project README.
6Context7
Upstash’s Context7 serves up-to-date, version-specific documentation for thousands of libraries, which reduces answers based on outdated APIs. Project page: github.com/upstash/context7.
claude mcp add context7 -- npx -y @upstash/context7-mcp
Best for
working with frameworks that change quickly.
Keep in mind
free use has rate limits; an API key from Context7 raises them.
7Terraform
HashiCorp’s server looks up providers, resources and modules in the Terraform Registry, so the agent writes configuration against current provider documentation. Project page: github.com/hashicorp/terraform-mcp-server.
claude mcp add terraform -- docker run -i --rm hashicorp/terraform-mcp-server
Best for
writing and upgrading Terraform code.
Keep in mind
it needs Docker; it reads registry documentation and doesn’t run plan or apply for you.
8AWS Documentation
AWS Labs maintains a collection of MCP servers. The documentation server searches and reads official AWS docs; others in the collection cover services such as CloudWatch, ECS and pricing. Project page: github.com/awslabs/mcp.
claude mcp add aws-docs -- uvx awslabs.aws-documentation-mcp-server@latest
Best for
getting accurate answers about AWS services and CLI options.
Keep in mind
servers that touch your account use your AWS credentials; give them a read-only IAM role.
9Grafana
Grafana’s official server lets the agent search dashboards, query Prometheus and Loki data sources and look at alerts and incidents. Project page: github.com/grafana/mcp-grafana.
claude mcp add grafana -- mcp-grafana
Best for
investigating incidents: “show error rates for the API since 10:00 and the matching logs”.
Keep in mind
install the binary from the project’s releases and set your Grafana URL and a service account token as described in the README; use a Viewer-role token.
10Memory
A simple knowledge-graph store that lets the agent remember facts between sessions, such as server names, conventions and past decisions. Project page: github.com/modelcontextprotocol/servers.
claude mcp add memory -- npx -y @modelcontextprotocol/server-memory
Best for
keeping context across long-running projects.
Keep in mind
everything is stored in a local file in plain text, so don’t let it record secrets.
Manage Your MCP Servers
In Claude Code, these commands list, inspect and remove servers. Add –scope project when adding a server to save it in the repository’s .mcp.json file and share it with your team:
claude mcp listclaude mcp get githubclaude mcp remove github
New to terminal agents? Start with installing Claude Code or Codex CLI, then add Git, GitHub and Fetch. They cover most daily work.
How did it make you feel?
