MCP Servers on Vibe Coding

Filesystem

Mon, 01 Jan 0001 00:00:00 +0000

The Filesystem server exposes a sandboxed slice of your local disk to an agent — reading, writing, listing, and searching files within directories you explicitly allow.

Why it matters

It’s one of the original reference MCP servers, and the clearest example of the protocol’s core promise: instead of every agent vendor writing its own file-access layer, any MCP-compatible client can use this one server.

GitHub

Mon, 01 Jan 0001 00:00:00 +0000

The GitHub MCP server lets an agent create and comment on issues, open pull requests, search code and repositories, and manage branches — all through GitHub’s API instead of shelling out to git and gh directly.

Postgres

Mon, 01 Jan 0001 00:00:00 +0000

The Postgres server gives an agent introspection into a database’s schema — tables, columns, relationships — and the ability to run read-only queries against it.

Why it matters

Without this, an agent writing SQL is guessing at your schema from memory or from whatever you paste in. With direct, read-only access, it can verify a query actually runs and returns the shape of data you expect before you ever run it yourself.

Slack

Mon, 01 Jan 0001 00:00:00 +0000

The Slack server connects an agent to a workspace — listing channels, reading recent messages, searching history, and posting replies — using a bot token scoped to whatever permissions you grant it.

Puppeteer

Mon, 01 Jan 0001 00:00:00 +0000

The Puppeteer server gives an agent control of a real (or headless) Chrome browser — navigating to URLs, clicking elements, filling in forms, and capturing screenshots as it goes.

Why it matters

Some tasks can only be verified by actually looking at rendered output — a UI change, a broken layout, a JavaScript-heavy page that a plain HTTP fetch can’t render. Browser automation closes that verification gap.

Brave Search

Mon, 01 Jan 0001 00:00:00 +0000

The Brave Search server gives an agent access to web and local search results through Brave’s Search API, without routing through another AI vendor’s index.

Why it matters

Models have a training cutoff and no innate awareness of anything after it. A search connector lets an agent verify current information — a library’s latest version, a recent API change, a live status — instead of confidently guessing from stale training data.

Memory

Mon, 01 Jan 0001 00:00:00 +0000

The Memory server stores facts as a local knowledge graph — entities, relationships, observations — that persists between sessions, giving an agent a lightweight, inspectable form of long-term memory.

Why it matters

Without persistence, every new conversation starts with zero context about you, your preferences, or past decisions. A local, file-backed memory store closes that gap without relying on a specific vendor’s built-in memory feature.

Google Drive

Mon, 01 Jan 0001 00:00:00 +0000

The Google Drive server lets an agent search your Drive and read the contents of documents and spreadsheets, using OAuth scoped to your account.

Why it matters

A lot of real work context — specs, meeting notes, planning docs — lives in Docs and Sheets, not in a codebase. This connector lets an agent reference that material directly instead of you manually copying content into a prompt.

Sentry

Mon, 01 Jan 0001 00:00:00 +0000

The Sentry server gives an agent access to issues, stack traces, and event details from a Sentry project, so it can investigate a production error with the same data a human engineer would open the dashboard to see.

Playwright

Mon, 01 Jan 0001 00:00:00 +0000

Playwright MCP gives an agent structured control over a real browser — Chromium, Firefox, or WebKit — using Playwright’s accessibility tree instead of pixel-based screenshots. It exposes tools for navigation, clicking, form filling, and page inspection so an agent can complete multi-step web tasks deterministically.

Chrome DevTools

Mon, 01 Jan 0001 00:00:00 +0000

Chrome DevTools MCP connects a coding agent to a live Chrome instance, exposing the same network, console, and performance-tracing tools available in DevTools. It lets an agent navigate pages, inspect requests, read console errors, and record performance traces from inside its own workflow.

Notion

Mon, 01 Jan 0001 00:00:00 +0000

Notion MCP exposes a workspace’s pages, databases, and blocks to an agent through Notion’s API, so it can search, read, and edit content without a human copying text back and forth. It’s available both as a locally run npm package and a hosted remote endpoint.

Linear

Mon, 01 Jan 0001 00:00:00 +0000

Linear MCP is Linear’s own hosted server, reachable at mcp.linear.app, that lets an agent create, update, and search issues, projects, and cycles in natural language instead of through the UI. It authenticates via OAuth on first connection rather than a static API token.

Figma

Mon, 01 Jan 0001 00:00:00 +0000

Figma’s Dev Mode MCP server exposes a selected frame or component’s structure, styles, and metadata directly to a coding agent, from either the Figma desktop app or a hosted endpoint. It lets an agent generate frontend code grounded in the real design file instead of a screenshot or description of it.

Stripe

Mon, 01 Jan 0001 00:00:00 +0000

Stripe’s official MCP server exposes its payments API — customers, charges, subscriptions, refunds, invoices — as tools an agent can call directly, available as a local npx package or Stripe’s hosted remote endpoint. It’s part of Stripe’s broader agent toolkit for LLM framework integration.

Cloudflare

Mon, 01 Jan 0001 00:00:00 +0000

Cloudflare’s official MCP server gives an agent access to a Cloudflare account — deploying Workers, querying D1 databases, managing KV namespaces and R2 buckets, and more — through a single hosted endpoint using code-execution-style tools.

Context7

Mon, 01 Jan 0001 00:00:00 +0000

Context7 resolves a library name (like “Next.js” or “shadcn/ui”) to a version-specific ID, then pulls current documentation and code snippets straight from the source into an agent’s prompt. It works with no signup for basic usage, with a free API key available for higher rate limits.

Sequential Thinking

Mon, 01 Jan 0001 00:00:00 +0000

The Sequential Thinking server provides a single tool that lets an agent externalize a chain of reasoning as discrete, revisable steps rather than reasoning silently in one pass. The host model decides when to call it, and can branch, revise, or extend earlier steps as its understanding develops.

Fetch

Mon, 01 Jan 0001 00:00:00 +0000

The Fetch server retrieves a given URL and converts its HTML into clean markdown sized for an LLM context window, with a start-index parameter so an agent can page through long documents in chunks. It’s one of the official reference servers in the modelcontextprotocol/servers repo.

Redis

Mon, 01 Jan 0001 00:00:00 +0000

Redis’s official MCP server lets an agent read and write Redis data structures — strings, hashes, lists, sets, streams — and run vector similarity search, using natural language instead of raw Redis commands. It connects to any reachable Redis instance via a connection URL.

Supabase

Mon, 01 Jan 0001 00:00:00 +0000

Supabase’s official MCP server gives an agent access to a Supabase project’s Postgres schema and data, Edge Functions, and project configuration, with a read-only mode for safer exploratory use. It’s built and maintained directly by the Supabase team.