Best MCP Servers for Gemini CLI in 2026

Gemini CLI's MCP onboarding is genuinely lower-friction than most clients in this series: a gemini mcp add command writes your settings.json for you, no hand-editing required. Firecrawl published its own Gemini CLI buyer's guide on August 17, 2026; this one runs the same comparison with Crawlora in it, sourced from each vendor's own docs and Gemini CLI's own documentation, not paraphrased from a listicle.

Gemini CLI's MCP config

Gemini CLI reads MCP server definitions from a JSON file at ~/.gemini/settings.json (user-level, applies everywhere) or .gemini/settings.json at your project root (project-scoped). The root key is mcpServers, the same shape Claude Desktop uses.

Gemini CLI supports two server shapes, and they take different fields:

• Command-based (local/stdio): { "command": "...", "args": [...], "env": {...}, "cwd": "...", "timeout": ..., "trust": ..., "includeTools": [...], "excludeTools": [...] } — runs a local process, typically via npx.
• URL-based (remote): either url (an SSE endpoint) or httpUrl (an HTTP-streaming endpoint) — pick the one your server actually speaks, they're not interchangeable — plus an optional headers object for auth:

"headers": {
  "Authorization": "Bearer your-api-token"
}

You don't have to hand-edit the file, either: gemini mcp add [options] <name> <commandOrUrl> [args...] writes the same JSON for you, with flags for -t/--transport (stdio/sse/http), -e/--env, -H/--header, --timeout, --trust, --include-tools, --exclude-tools. Manage servers afterward with gemini mcp list, gemini mcp remove <name>, and gemini mcp enable/disable <name>.

What to evaluate, in Gemini CLI specifically

• Transport field: Gemini CLI splits remote servers into url (SSE) and httpUrl (HTTP streaming). A server on the wrong field silently fails to connect, so check which transport each vendor speaks before copying a config from another client.
• CLI add versus hand-edit: gemini mcp add writes settings.json for you and takes -e for env and -H for headers. Prefer it for anything with a secret, so the key never goes through your editor's history.
• Tool filtering: includeTools and excludeTools are per-server settings. A server with 1,398 tools is fine if you scope it to the dozen your project uses; a server with 60 that all need a browser is not automatically lighter.
• Generic crawl (any URL → markdown) versus structured platform data (named sites → JSON). Gemini CLI agents do both in one session; plan for two servers, not one.
• Billing on failure: a blocked fetch inside a Gemini CLI loop retries. Whether each attempt is billed changes the cost of a long session.

The contenders

Firecrawl MCP: the vendor that wrote the first Gemini CLI guide

Firecrawl's MCP server exposes scrape, crawl, map, search, extract and interact tools against any public URL. Firecrawl published its own Gemini CLI quickstart and buyer's guide on August 17, 2026, and its recommended config is the standard command-based shape: npx -y firecrawl-mcp with FIRECRAWL_API_KEY under env, confirmed with /mcp list. Or in one line:

gemini mcp add -e FIRECRAWL_API_KEY=$FIRECRAWL_API_KEY firecrawl npx -y firecrawl-mcp

Choose it when the session's web calls are "read this page or this repo's docs into context", which is most of them for a coding agent.

Bright Data: the same local-process shape, with Pro mode

Bright Data publishes its own Gemini CLI guide with the identical command-based config: npx -y @brightdata/mcp, API_TOKEN under env, and "PRO_MODE": "true" to unlock the full 60+-tool set (the default is the smaller free Rapid tier). Coverage spans named e-commerce, social, business and finance platforms plus proxy and browser infrastructure. Because everything runs through a local npx process, excludeTools is worth setting: the browser-automation tools are rarely what a Gemini CLI coding session needs. Choose it when anti-bot reliability on hard, named targets is the constraint.

Zyte: not currently available for Gemini CLI

Worth stating plainly rather than skipping: Zyte's two Claude-adjacent products (the Claude Code plugin and the GitHub-Copilot-specific Web Scraping Copilot VS Code extension) are each locked to one client, and neither connects to Gemini CLI. If your stack is Gemini-first, Zyte is not in the running here.

Crawlora: httpUrl, not url, and nothing to run

Crawlora's MCP server connects to Gemini CLI as an httpUrl-based remote server, not url. Crawlora's hosted endpoint speaks Streamable HTTP, and Gemini CLI reserves url for SSE, so a config copied from a client that uses a single url field will not connect here. The CLI shortcut sets the right transport for you:

gemini mcp add --transport http crawlora https://mcp.crawlora.net/mcp \
  --header "Authorization: Bearer $CRAWLORA_API_KEY"

No npx process to install or keep running, on your machine or a remote one. 1,398 tools map one-to-one onto documented REST endpoints across search, maps, e-commerce, social, finance, travel, jobs, media, app stores and reviews; use includeTools to scope a project to the handful it needs. Responses are normalized JSON, billed only on success, with 2,000 free credits a month and no card. Choose it when the agent needs a record from a named platform (a price, a ranking, a listing, a filing) rather than a page to read.

How to choose, by job

• "Read this page / these docs into context" → Firecrawl MCP.
• "Get me the data off this hard, named site, at volume" → Bright Data, with excludeTools trimming the browser tools.
• "Give me the price / ranking / listing / filing from a known platform, as JSON, with nothing installed" → Crawlora MCP, on httpUrl.
• "I want Zyte's Scrapy-generation workflow" → Claude Code or GitHub Copilot for now, not Gemini CLI.
• Config copied from another client won't connect → check url versus httpUrl first; it is the most common Gemini CLI MCP mistake.

Add structured web data to Gemini CLI

1,398 tools across 155 platform groups, one API key, billed only on success. 2,000 free credits a month, no card.

Set up Crawlora MCPBrowse the API catalog

Sources

Related reading

• Best MCP Servers for Warp in 2026: another CLI-first client, same three vendors.
• Best MCP Servers for JetBrains IDEs in 2026: the header-auth-bridge gotcha for a different client.
• Best MCP Servers for GitHub Copilot in 2026.
• Best MCP Servers for Windsurf in 2026.
• Best Web Scraping MCP Servers & Skills for Claude Code in 2026.
• Crawlora vs Firecrawl · Crawlora vs Bright Data.
• Give your AI agent live web data with MCP.