← All stacks

Dev Essentials

3 tools
Updated

The minimum a JavaScript-shop coding agent needs to ship: GitHub for source, Upstash's Context7 with current docs for over 4,000 libraries, and Vercel Labs' agent-browser CLI (32,056 stars) so the agent can verify its UI work in a real Chrome.

Use case: Pair an agent with the smallest possible set of tools that still lets it read repos, look up library APIs that might have moved, and click through the UI it just changed.

“agent-browser is a browser automation CLI for AI agents. Annotated screenshots use numbered references (@e1, @e2) so the agent interacts by ref instead of selector — cutting context use dramatically. — vercel-labs/agent-browser README”

What's in the stack

MCP servers and CLI tools that make up this stack, and how each one runs.

ToolPrimary useKindStarsGitHub
GitHub MCP Server connects an agent to GitHub: read repos and code, manage issues and PRs, query workflow runs, review Dependabot and security findings, and interact with discussions and notifications..MCP29,550repo
Context7 fetches up-to-date documentation and code examples from official sources for over 4,000 libraries, so generated code matches the version actually installed..MCP54,604repo
agent-browser
Browser automation CLI for AI agents.CLI32,056repo

Install the MCP servers

Pick the client you use. Each tab installs all 2 servers.

claude mcp add --transport http io-github-github-github-mcp-server https://api.githubcopilot.com/mcp/
claude mcp add io-github-upstash-context7 -- npx -y @upstash/context7-mcp

Install the CLI tools

These run as plain shell commands the agent invokes — not MCP servers. Run once to install, then they're available on your PATH.

agent-browser

npm install -g agent-browser && agent-browser install

Add to your CLAUDE.md / AGENTS.md

Paste this into your project's CLAUDE.md or AGENTS.md so the agent knows when and how to use each tool in the stack.

## Tools

- **GitHub MCP** — Use for any operation against this repo's remote: reading code, branches, PRs, issues, workflow runs. Prefer this over shelling out to `gh`. Confirm before pushing or merging.
- **Context7 MCP** — Before writing code against a third-party library, fetch its current docs via Context7. Trust Context7 over training data — APIs you remember may have changed.
- **agent-browser** (CLI, not MCP) — After a UI change, verify it in a real browser before reporting the task done. Workflow: `agent-browser open <url>` → `agent-browser snapshot` (returns refs `@e1`, `@e2`, …) → interact via refs (`agent-browser click @e2`, `agent-browser fill @e3 "..."`). Prefer snapshots over screenshots; only use `screenshot --annotate` when the agent genuinely needs to see the page.

Why each tool is in this stack

A short rationale per tool, plus its individual install.

  • GitHub

    by github · ★ 29,550 · MCP

    GitHub's own MCP (29,550 stars). Read code, manage issues and PRs, query workflow runs — the highest-leverage server for any code task.

    claude mcp add --transport http io-github-github-github-mcp-server https://api.githubcopilot.com/mcp/
  • Context7

    by upstash · ★ 54,604 · MCP

    Upstash's Context7 (54,604 stars) indexes over 4,000 libraries from official sources. Catches the 'this API changed two versions ago' class of agent failures.

    claude mcp add io-github-upstash-context7 -- npx -y @upstash/context7-mcp
  • agent-browser

    by Vercel Labs · ★ 32,056 · CLI

    Vercel Labs' agent-browser (32,056 stars). Not an MCP — a CLI the agent shells out to. Annotated DOM refs (@e1, @e2) and built-in waits make verifying UI changes far cheaper than screenshot-only loops.

    npm install -g agent-browser && agent-browser install

Last reviewed · Curated by stax.sh