Primary keyword
mcp server list for claude code
This page captures research intent before the reader is ready to open a calculator.
AI agent directory
MCP Server List for Claude Code, Codex, and VS Code Agents
Most MCP server lists are broad. This page is built for one searcher: a developer deciding which servers to connect to a coding agent without giving every tool unnecessary access.
Separates coding-agent use cases from generic MCP showcases.
Labels local access, auth, hosted dependency, and safer first tests.
Designed to be readable by humans, search engines, and AI agents.
Audience
Developers using Claude Code, Codex, VS Code agents, or local coding assistants who need a short list before wiring tools into a project.
Each guide is designed to hand the reader off to the right calculator.
Core guide
A practical MCP server directory for coding agents, filtered by task, local risk, auth requirement, and whether the server fits Claude Code, Codex, or VS Code workflows.
Start with the task, not the protocol name
MCP is useful only when the server gives an agent a tool it truly needs. For coding agents, the first split should be filesystem, GitHub, docs/search, browser, database, project management, or cloud deployment.
Use the lowest-risk server first
A local coding workflow should begin with read-only or narrowly scoped access. Avoid adding credentials, write access, or broad browser control until the exact workflow has proved useful.
- Prefer read-only docs, search, or repository context for the first test.
- Add GitHub or database access only after permissions and audit trail are clear.
- Keep production credentials out of local experiments.
When a list page still earns the click
A short AI summary can define MCP, but it cannot safely choose a server for a real repo. The click is earned by filters, risk labels, examples, and a repeatable selection checklist.
Need a quick build-or-skip decision for one AI workflow?
Request a $49 WhatsApp AI agent or n8n workflow audit. Send one use case and get the smallest useful stack, template, or skip recommendation.
MCP server selection matrix for coding agents
Filesystem or repo context
Best first use
Let the agent inspect local files or a bounded workspace before suggesting edits.
Risk label
Medium: local file access must be scoped to the project.
Good fit
Claude Code, Codex, VS Code agents.
GitHub issues and pull requests
Best first use
Summarize issues, read PR context, or prepare review notes before code changes.
Risk label
Medium-high: requires account auth and repository permissions.
Good fit
Coding agents that already work from branches or PRs.
Docs and web search
Best first use
Fetch official docs or current package guidance when model memory may be stale.
Risk label
Low-medium: outbound requests and source quality still matter.
Good fit
Any coding agent doing dependency or API work.
Database or analytics
Best first use
Inspect schema, run bounded queries, or explain metrics before changing application logic.
Risk label
High: data access, PII, and write permissions must be controlled.
Good fit
Internal tools and analytics-heavy apps.
Browser automation
Best first use
Verify UI behavior, screenshots, and local app flows after code changes.
Risk label
Medium-high: browser control should avoid logged-in account risk unless needed.
Good fit
Frontend apps, QA workflows, and local dev servers.
Project management
Best first use
Read task context, acceptance criteria, and implementation notes from external systems.
Risk label
Medium: usually account-authenticated but less dangerous than production data access.
Good fit
Teams where coding tasks live outside GitHub.
FAQ
Is this a complete list of every MCP server?
No. It is a coding-agent selection list. The goal is to choose a safe first server for a practical workflow, not to mirror every public repository.
Should I connect all useful MCP servers at once?
No. Add one server, run one workflow, then decide whether the extra tool access was worth the risk and setup cost.
Does this apply to both Claude Code and Codex?
Yes. The exact client setup differs, but the selection logic is similar: task fit, permission scope, credential risk, and whether the server reduces real manual work.
Where does the paid template fit?
Use the template when you need to turn the directory into an implementation checklist for one repo or one agent workflow.