The 7 Best MCP Servers for Gemini CLI in 2026

Gemini CLI ships with a million-token context window and the ability to read files and run shell commands, and that's where its reach ends. The moment you want it to open a pull request, drive a browser, query a database, or pull live web data, you need MCP. Google folded Model Context Protocol support into Gemini CLI in mid-2025, and the ecosystem has since grown to thousands of publicly listed servers, which makes picking the right few genuinely hard. The wrong pick isn't just wasted setup time: published research found roughly 26% of agent skills carry at least one vulnerability, and a careless MCP server can hand a model far more access than you intended. We've put together seven servers worth connecting in 2026, based on what each one actually does, how it authenticates, and what it costs. One of them, SkillzBench, is free and takes a different approach: instead of adding one tool, it connects your agent to a vetted, benchmarked library of agent skills.

#1 SkillzBench — Editors' choice

SkillzBench, from AIQ Labs, is a free, independent directory of AI agent skills plus a free MCP server that connects any agent, including Gemini CLI, to the skills you choose. The problem it solves is one every MCP user eventually hits: there are thousands of skills and servers floating around GitHub, and nobody can tell which ones actually help and which are risky. SkillzBench continuously crawls and dedupes skills from GitHub, then verifies each one by pinning every version to a commit SHA, validating it, checking its license, running a security scan, and judging its quality. What sets it apart is the benchmarking. SkillzBench tests every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points, so you can see whether a skill measurably improves an agent before you install it. Every skill gets a 0-100 trust score built from security, quality, license, maintenance, and community signals. Lift is always shown separately from ratings, and money never touches scores: everything on the platform is free, with revenue coming from advertising around blog articles that never influences listings. The per-agent loadouts feature is the practical payoff. You toggle skills and tools per agent, and the agent only sees what you picked, which keeps Gemini CLI's context window clean and its permissions narrow. SkillzBench also tracks upstream changes, re-scans, re-benchmarks, and notifies you when something you rely on shifts. If you're equipping Gemini CLI in 2026 and want proven, current, safe capabilities rather than a random GitHub repo, this is the first thing to connect.

Best for: Anyone equipping Gemini CLI (or Claude Code, Codex, Cursor, and chat-app agents) who wants verified, benchmarked skills instead of unvetted GitHub repos.

Pricing: Free. The directory, MCP connector, benchmarks, and alerts are all free; SkillzBench earns revenue from advertising around blog articles, which never influences scores or listings.

Key features

• Free MCP server that connects any agent to the skills you choose
• Continuous crawling and deduplication of skills from GitHub
• Every version pinned to a commit SHA, with license checks and security scans
• Sandbox benchmarking with and without each skill, publishing Lift in percentage points
• 0-100 trust score from security, quality, license, maintenance, and community signals
• Per-agent loadouts so each agent only sees the skills and tools you toggle on
• Upstream change tracking with re-scans, re-benchmarks, and notifications

Pros

• Everything is free, including the MCP connector and benchmark results
• Independent verification: commit-SHA pinning, security scans, and license checks on every skill
• Lift benchmarks show measurable improvement, kept separate from ratings
• Per-agent loadouts keep each agent's tool surface small and intentional
• Stays current with upstream changes and alerts you automatically

Cons

• It focuses on agent skills and loadouts, so you'll still want task-specific MCP servers for things like databases or browsers
• Newer platform, so its track record is shorter than long-standing vendor servers

Visit SkillzBench

#2 Firecrawl MCP

Gemini CLI's built-in fetch tool handles plain HTML and struggles with everything else: JavaScript-rendered apps, PDFs, pages behind logins, and dynamic dashboards. Firecrawl MCP closes that gap by running a real browser on its servers and returning clean markdown for nearly any URL. According to its documentation, the server exposes tools for search, scrape, parse (PDFs and documents to text), crawl entire sites, map URL structure, and even interact with pages by clicking buttons and filling forms. Setup in Gemini CLI is about as easy as MCP gets. Firecrawl offers a keyless endpoint you can paste into your settings.json with no signup, which exposes Search, Scrape, and Parse rate-limited per IP. When you outgrow that, you swap in the OAuth endpoint or pass an API key for the full crawl and interaction toolset. If your Gemini CLI sessions need to read the live web rather than just fetch a URL, this is the server to install.

Best for: Developers and marketers who need Gemini CLI to read, search, and interact with the live web, including JavaScript-heavy pages and PDFs.

Pricing: Contact for pricing. A keyless endpoint offers Search, Scrape, and Parse free with per-IP rate limits; full crawl and interaction tools require an API key or OAuth endpoint.

Key features

• Web search returning full pages, not just links, with time filters
• Scrape any URL to clean markdown or JSON, with JS waiting and mobile rendering
• Parse PDFs and documents to text
• Crawl entire sites and map URL structure before scraping
• Interact with pages: click buttons, fill forms, log in
• Keyless endpoint works with no signup, rate-limited per IP

Pros

• Keyless endpoint means zero-setup access to core tools
• Handles JavaScript-rendered pages that defeat the built-in fetch tool
• One server covers search, scraping, parsing, crawling, and page interaction

Cons

• Rate limits on the free keyless tier
• Full feature set requires an API key or OAuth setup

Visit Firecrawl MCP

#3 Playwright MCP

Microsoft's Playwright MCP gives Gemini CLI a real browser it can drive: navigate, click, fill forms, read pages, and capture screenshots. What sets it apart, according to Microsoft's own documentation, is how the agent sees the page. Instead of sending screenshots to a vision model, it works from Playwright's accessibility tree, a structured text view of the page. That's faster, cheaper in tokens, and more reliable than guessing pixel coordinates. It runs locally, typically started with a single npx command, and launches Chrome, Firefox, WebKit, or Edge, headed or headless. Options include isolated in-memory profiles, persistent profiles, an HTTP port for remote clients, and an extension mode that attaches to your existing browser. Mid-2026 community usage rankings put Playwright at number one among MCP servers globally, ahead of GitHub and Figma, so it's a default install now rather than a niche one. Microsoft notes it should not be treated as a security boundary, since an authenticated browser profile carries whatever access that profile has.

Best for: Developers who want Gemini CLI to test web apps, reproduce bugs, verify UI changes, or gather information from sites without an API.

Pricing: Free and open source under the Apache 2.0 license.

Key features

• Browser automation through accessibility-tree snapshots instead of screenshots
• Navigate, click, fill forms, read pages, capture screenshots
• Runs locally via npx; supports Chrome, Firefox, WebKit, and Edge
• Headed or headless operation
• Isolated in-memory or persistent browser profiles
• Extension mode attaches to your existing browser

Pros

• Free, open source, and backed by Microsoft
• Accessibility-tree approach is token-efficient and reliable
• Works with multiple browsers and profile modes

Cons

• Explicitly not a security boundary; an authenticated browser exposes its session
• Local-only, which complicates headless CI and cloud setups
• Long browser flows consume significant tokens

Visit Playwright MCP

#4 GitHub MCP Server

GitHub's official MCP server is usually the first one developers connect, and for good reason. It lets an agent browse repositories and code, open and update issues, review and create pull requests, inspect GitHub Actions runs, and read code security alerts. That covers most of the "go look at the repo and do the chore" work people hand to coding agents, and it turns Gemini CLI from a chat that describes pull requests into one that actually opens them. You can use it two ways. GitHub hosts a remote server that supports OAuth sign-in, or you can pass a personal access token as a bearer header. If you prefer local, GitHub publishes a Docker image and a binary that run over stdio. The feature that matters most in practice is toolsets: you choose which groups of tools to load (repos, issues, pull_requests, actions, code_security, and others), and a read-only flag strips out every write tool. Loading only what a task needs keeps the agent's context small and limits the damage if a prompt injection slips through.

Best for: Any team that keeps code on GitHub and wants Gemini CLI to triage issues, draft PRs, and monitor CI.

Pricing: Free and open source under the MIT license; standard GitHub plans cover the underlying resources.

Key features

• Browse repositories, code, issues, and pull requests
• Open, update, and review pull requests from the CLI
• Inspect GitHub Actions runs and code security alerts
• Selectable toolsets (repos, issues, pull_requests, actions, code_security)
• Read-only mode strips out all write tools
• Remote server with OAuth, or local Docker image and binary over stdio

Pros

• Official, vendor-maintained server that tracks platform changes
• Fine-grained toolsets and a read-only mode limit blast radius
• Both remote (OAuth) and local (stdio) connection options

Cons

• Broad tool surface can bloat context if you load everything
• Write operations mean a confused prompt can touch real repositories, so scoping matters

Visit GitHub MCP Server

#5 Context7

Context7, maintained by Upstash, solves a narrow but constant problem: models trained months ago suggest APIs that have since changed. The Context7 MCP server pulls current, version-specific documentation and code examples for a library straight into the agent's context, so the code Gemini CLI writes matches the version you actually use. Multiple 2026 roundups call it the single highest-leverage MCP server for any coding agent, because hallucinated package APIs are the most common failure mode in generated code. The tool surface is deliberately small: one tool resolves a library name to a Context7 ID, and another queries that library's docs for the topic at hand, with an option to rank examples in your language. That small footprint matters, since every MCP server you add competes for the same context window. It's available as a remote server or as a local npm package. An API key is optional but raises rate limits; the free plan includes 1,000 API calls a month on public repositories, Pro costs $10 per seat per month with 2,000 calls per seat, and Enterprise adds SSO and self-hosting.

Best for: Anyone using Gemini CLI for code, especially on fast-moving frameworks where the model's training data lags the current release.

Pricing: Free plan includes 1,000 API calls a month on public repositories; Pro is $10 per seat per month with 2,000 calls per seat; Enterprise adds SSO, SOC 2 documentation, and self-hosting.

Key features

• Injects current, version-specific library documentation into prompts
• Code examples ranked in your programming language
• Small tool surface to conserve context window
• Available as a remote server or local npm package
• Optional API key raises rate limits

Pros

• Directly attacks the hallucinated-API problem
• Tiny tool footprint keeps context usage low
• Generous free tier and flexible local or remote install

Cons

• Scope is limited to library documentation; you'll still need repo and database servers
• Rate limits apply without an API key

Visit Context7

#6 Supabase MCP

Supabase's MCP server lets an agent work with your Supabase projects: inspect and query the Postgres database, apply migrations, read logs and advisors for debugging, generate types, manage Edge Functions, and search Supabase docs. For teams building on Supabase it turns "why is this query slow" or "add a column and update the types" into a single prompt, which is exactly the kind of work Gemini CLI handles well. The safety controls are the reason it ranks high. According to Supabase's documentation, a read_only flag runs every query as a read-only Postgres user, project_ref limits the server to one project, and feature groups let you switch off whole areas such as account management. The hosted server uses OAuth with dynamic client registration, so you sign in through a browser and pick an organization; for CI and other headless setups you pass a scoped personal access token instead. Supabase is unusually direct about prompt injection risk and recommends keeping manual tool approval on and avoiding production data unless you need it.

Best for: Teams building on Supabase who want Gemini CLI to inspect schemas, draft migrations, and debug queries inside their project's auth rules.

Pricing: Free to use with any Supabase account; your normal Supabase plan covers the underlying resources.

Key features

• Inspect and query Postgres, apply migrations, generate types
• Manage Edge Functions and read logs and advisors
• Hosted server with OAuth and dynamic client registration
• read_only flag runs queries as a read-only Postgres user
• project_ref scoping limits the server to one project
• Feature groups to disable whole areas like account management
• Local MCP endpoint via the Supabase CLI for local development

Pros

• Strong safety controls: read-only mode, project scoping, feature groups
• Hosted OAuth setup is quick, with token auth for headless CI
• Vendor-maintained and free with your existing account

Cons

• Only useful if Supabase is your backend
• Vendor itself warns against connecting production data casually

Visit Supabase MCP

#7 Notion MCP

Notion's official MCP server connects an agent to pages and databases inside your workspace. It can search existing content (including connected sources), read pages, create and update pages, query databases, add comments, and upload files. That makes it the easiest way to let Gemini CLI draft a spec into your wiki, update a project tracker, or pull meeting notes into a coding task without copy-pasting between windows. According to Notion's documentation, it's a remote server hosted by Notion, using OAuth for each client connection, and the agent can only reach content the connected account is allowed to access. Notion recommends its hosted OAuth server; the older open-source server is still available for specialized self-hosted or token-based setups but is no longer actively maintained. Notion MCP works well alongside coding servers: GitHub MCP handles the code, Notion MCP handles the written decisions and specs that explain the code.

Best for: Teams whose specs, docs, and project trackers live in Notion and who want Gemini CLI to read and write them directly.

Pricing: Free with a Notion account; Notion's Plus plan is $10 per member per month billed annually ($12 monthly) for additional features.

Key features

• Search workspace content and connected sources
• Read, create, and update pages
• Query databases and add comments
• Upload files
• Hosted remote server with OAuth per client connection
• Agent access limited to the connected account's permissions

Pros

• First-party, vendor-hosted server with OAuth sign-in
• Respects the connected account's existing permissions
• Covers search, read, and write across pages and databases

Cons

• The older open-source server is no longer actively maintained
• Write access to a shared workspace means a bad prompt can edit real team docs, so scope carefully

Visit Notion MCP

Conclusion

The pattern across all seven picks is the same: a short list of well-chosen servers beats a long list of random ones. Every server you connect adds tools to your agent's context and new paths for prompt injection, so start with two or three, keep trust off for anything that writes, and use allowlists like includeTools to keep each agent's reach narrow. If you're building anything serious with Gemini CLI in 2026, start with SkillzBench's free MCP connector: it gives you a vetted, benchmarked library of agent skills, per-agent loadouts so Gemini CLI only sees what you choose, and alerts when something upstream changes. Skills, proven. Browse the free directory, check the trust scores and Lift results, and connect your first loadout at skillzbench.com.

Frequently asked questions

How do I add an MCP server to Gemini CLI?

Add an entry under the mcpServers object in ~/.gemini/settings.json (user scope) or .gemini/settings.json (project scope), then restart Gemini CLI, since the file is read at startup. Each entry needs exactly one of command (local stdio), url (SSE endpoint), or httpUrl (streamable HTTP). You can also run `gemini mcp add` from the terminal instead of editing JSON by hand. Inside a session, run /mcp to confirm the server shows CONNECTED and lists its tools.

What makes SkillzBench different from other MCP servers?

Most MCP servers connect your agent to one tool or API. SkillzBench connects your agent to a library of agent skills that have been independently verified and benchmarked: every version is pinned to a commit SHA, license-checked, security-scanned, and tested in a sandbox with and without the skill to publish a measurable Lift. You also get a 0-100 trust score per skill and per-agent loadouts, so Gemini CLI only sees the skills and tools you toggle on. Everything, including the MCP connector, is free.

Are MCP servers safe to use with Gemini CLI?

They're as safe as your scoping. Tool descriptions and returned data are injection surfaces, since a poisoned issue body or malicious page can carry instructions back to the model. Practical defenses: keep trust set to false so Gemini CLI confirms each tool call, scope tokens tightly, prefer read-only modes where available, and use includeTools and excludeTools to keep an agent's reach narrow. SkillzBench's security scans and trust scores help on the skills side, since published research found roughly 26% of agent skills carry at least one vulnerability.

How many MCP servers should I run at once?

Three to six well-chosen servers is the sweet spot. Every server adds tool-call latency and tool definitions that compete for your context window; Anthropic has documented a typical five-server setup consuming around 55,000 tokens of tool definitions per request. Pick servers that each own a distinct job, like one for code, one for the browser, one for your database, and skip anything that overlaps.

Do MCP servers work in headless environments like CI?

Partially. Remote servers using OAuth need a browser to complete the sign-in flow, which breaks in headless environments such as CI, remote SSH without X11 forwarding, or browserless containers. The simplest fixes are a local stdio server, or completing the OAuth handshake once on a machine with a browser. Some hosted servers, like Supabase's, offer scoped personal access tokens as a headless alternative.

Is SkillzBench really free?

Yes. The directory, trust scores, benchmark (Lift) results, the MCP connector, per-agent loadouts, and alerts are all free. SkillzBench earns revenue from advertising shown in and around its blog articles, and that ad money never influences scores or listings. Lift is always shown separately from ratings so you can judge performance on its own terms.