At a glance: google/mcp (4k stars, 448 forks, 30+ listed servers) + Gemini CLI (103k stars, 13.4k forks, native MCP client). Google provides the largest official MCP server catalog of any company — announced at Google Cloud Next ‘26 with 50+ managed and open-source MCP servers across GA and Preview. Combined with Gemini CLI, Deep Research agents with MCP support, and built-in MCP in Gemini API SDKs, Google has the most complete MCP ecosystem. Part of our AI Providers MCP category.
What changed (May 19, 2026 refresh): Google I/O 2026 (May 19) is consumer-focused — Gemini Intelligence layer for Android 17, Googlebook laptop category, Android XR glasses preview; new flagship Gemini model (Gemini 4.0 / Gemini Omni) expected but name unconfirmed at press time. Developer MCP content is from Google Cloud Next ‘26 (April) and subsequent announcements: Gemini API Docs MCP + Agent Skills deliver 96.3% pass rate on Gemini API evals with 63% fewer tokens vs. vanilla prompting. ADK 1.0 GA (Python, Go, TypeScript) with ADK for Java 1.0.0 (newly announced) now supports native MCP integration, Event Compaction, and Human-in-the-Loop. New Agent Registry centralizes agents and MCP servers. Agents CLI gives AI coding agents (Gemini CLI, Claude Code, Cursor) machine-readable access to the full agent stack. Genkit 2.0 adds native MCP server integration (TypeScript/Go/Dart/Python). A2A Protocol (Agent-to-Agent) launched as a companion to MCP for agent-to-agent communication. Project Mariner (browser AI agent) shut down May 4 — capabilities absorbed into Gemini API. Earlier in the month: Google Cloud Next ‘26 doubled the MCP server count from 24+ to 50+; Gemini CLI grew to 103k stars; Deep Research agents added MCP-powered private data analysis; Gemini Pro moved to paid-only (April 1).
Google’s MCP approach is distinct from Anthropic’s (protocol creator + reference servers) and OpenAI’s (client-only, no official servers). Google went wide with production-grade managed servers across their entire Cloud and Workspace portfolio, making it possible for AI agents to query BigQuery, navigate Google Maps, manage Kubernetes clusters, access Firestore, use Apigee APIs, analyze Looker dashboards, and work with Docs/Sheets/Gmail — all through standard MCP.
Alphabet/Google was founded in 1998 by Larry Page and Sergey Brin. As of early 2026: $402.8 billion annual revenue (FY 2025), approximately $2 trillion market capitalization, 190,820 employees, 3 billion+ Google Workspace users, and the Gemini model family powering AI across Search, Cloud, and developer tools. Key AI products include Gemini (3.1 Pro, 3 Flash, 2.5 series), Gemini CLI, Google AI Studio, Vertex AI, and Deep Research agents.
Architecture note: Google’s MCP strategy covers four layers: (1) managed remote servers — 50+ fully hosted by Google Cloud, requiring only authentication and a service endpoint (no separate MCP enablement needed since March 17); (2) open-source servers — self-hosted, covering Workspace apps and developer tools; (3) MCP client support — Gemini CLI, Deep Research agents, and Gemini API SDKs (Python/JavaScript) can consume any MCP server; (4) MCP + function calling — MCP tools and Google’s built-in function calling (Google Search grounding, code execution) can be combined in a single API request. Google is also a platinum member of the Agentic AI Foundation (AAIF) alongside AWS, Microsoft, Anthropic, and OpenAI.
What It Does
Managed Remote Servers (16+ listed, 50+ total across GA and Preview)
Fully hosted by Google Cloud — no infrastructure to manage. Since March 17, 2026, MCP endpoints are available by default when you enable a supported product (no separate MCP enablement needed). Announced at Google Cloud Next ‘26 with 50+ servers across GA and Preview:
Databases & Analytics
| Server | Status | What It Does |
|---|---|---|
| BigQuery | GA | Query enterprise data warehouses, interpret schemas, execute SQL — data stays in place and governed |
| AlloyDB for PostgreSQL | GA | PostgreSQL-compatible database with AI-optimized queries |
| Cloud SQL (MySQL/PostgreSQL/SQL Server) | GA | Natural language interaction with relational database fleets |
| Spanner | GA | Globally-distributed database with graph, relational, and semantic queries via SQL and GQL |
| Firestore | GA | Document database operations for serverless applications |
| Bigtable | GA | High-performance NoSQL for analytics and time-series workloads |
| Database Center | GA | Cross-database fleet management and diagnostics |
| Managed Service for Apache Kafka | GA | Create and manage Kafka clusters, topics, consumer groups, connectors, and ACLs |
| Managed Service for Apache Spark | — | Large-scale data processing |
| Looker | Available | Business intelligence dashboards and data exploration — accessible via MCP Toolbox from Gemini CLI, Claude, Cursor |
Infrastructure & Services
| Server | Status | What It Does |
|---|---|---|
| Compute Engine (GCE) | GA | Manage VMs, managed instance groups, disks, and snapshots |
| Kubernetes Engine (GKE) | GA | Container orchestration, cluster management, and deployment |
| Cloud Run | GA | Serverless container deployment and management |
| Cloud Storage | GA | Object storage operations (buckets, files, access control) |
| Cloud Resource Manager | GA | Project and resource organization across Google Cloud |
| Google Maps (Grounding Lite) | GA | Geocoding, directions, place search, and route validation |
| Google Security Operations (Chronicle) | GA | Security event analysis and threat investigation |
| Developer Knowledge API | GA | Connect IDEs and agents to Google’s documentation |
| Apigee | GA | Transform APIs into AI-ready tools using OpenAPI specs — no local MCP servers needed |
| Network Management API | Available | Connectivity tests and network diagnostics |
| Android Management API | Available | Manage Android devices, applications, and enterprise policies |
| Google Pay & Wallet | Available | Integrate payments and digital passes into agentic workflows |
| Pub/Sub | Announced | Event-driven messaging and proactive system alerts |
AI & Research
| Server | What It Does |
|---|---|
| Deep Research (deep-research-preview-04-2026) | Autonomous research agent — fast, for interactive surfaces. Can use MCP servers to query private databases, internal docs, and third-party data |
| Deep Research Max (deep-research-max-preview-04-2026) | Maximum comprehensiveness research agent — for background/async workflows. Extended test-time compute with iterative reasoning |
| Gemini API Docs MCP | Connects coding agents to live Gemini API documentation, SDKs, and model info via MCP — prevents hallucination of outdated API calls. Achieves 96.3% pass rate on Gemini API eval set with 63% fewer tokens per correct answer vs. vanilla prompting when combined with Agent Skills |
Agent Registry
Google announced an Agent Registry — a centralized directory for agents, MCP servers, and tools across the Google Cloud agent ecosystem. Works alongside the Google Official Skills Repository (Markdown-based compact agent instructions).
Open-Source Servers (15)
Self-hosted servers covering Workspace productivity, security, and developer tooling:
Workspace & Productivity
| Server | What It Does |
|---|---|
| Google Workspace | Docs, Sheets, Slides, Calendar, Gmail integration |
| Google Analytics | Website and app analytics data access |
Developer & Infrastructure
| Server | What It Does |
|---|---|
| Firebase | App platform — database, auth, hosting, cloud functions |
| Cloud Run (Gemini CLI Extension) | Serverless container deployment and management |
| Google Cloud Storage | Object storage operations (buckets, files, access control) |
| gcloud CLI | Bridge to the full Google Cloud CLI for any gcloud command |
| Google Cloud Observability | Monitoring, logging, and tracing across Cloud services |
| MCP Toolbox for Databases | Unified database access layer for building database MCP servers |
Security (NEW — google/mcp-security)
| Server | What It Does |
|---|---|
| Google Security Operations (Chronicle) | Threat detection, investigation, and hunting |
| Security Operations SOAR | Security orchestration, automation, and response |
| Google Threat Intelligence (GTI) | Access Google’s threat intelligence data |
| Security Command Center (SCC) | Cloud security and risk management |
AI & Creative
| Server | What It Does |
|---|---|
| Genmedia | Image generation (Imagen) and video generation (Veo) |
| Flutter/Dart | Mobile and cross-platform app development tools |
| Chrome DevTools | Browser debugging and performance analysis |
| Go | Go language development tools |
| Google Maps Platform Code Assist | Maps development toolkit |
Planned Servers (Coming Soon)
Google has announced upcoming MCP support for: Database Migration Service, BigQuery Migration Service, Memorystore, and Pub/Sub (managed servers not yet GA).
Google as MCP Client
Google provides MCP client support across multiple products:
Gemini CLI
- google-gemini/gemini-cli — 103k stars, 13.4k forks, 6,005 commits, Apache 2.0
- Open-source terminal AI agent with native MCP server support
- v0.40.1 (May 1, 2026) — latest stable; v0.41.0-preview.0 previews real-time voice mode
- v0.40.0 (April 28) — NEW: MCP resource listing and reading tools, bundled ripgrep for offline search, colorblind-friendly themes, prompt-driven four-tier memory system replacing legacy MemoryManagerAgent
- v0.39.0 (April 23) — /memory inbox command, MCP resource tools, advanced memory management
- Configure MCP servers in
~/.gemini/settings.json - Supports stdio and SSE transports
- Built-in tools: file operations, shell commands, web fetch, Google Search grounding
- Three release tracks: Nightly (daily), Preview (weekly), Stable (weekly)
- Install via npm, Homebrew, MacPorts, Anaconda, or Docker
- IDE stdio override security fix (RCE prevention) in v0.40.0
Deep Research Agents (NEW — April 21, 2026)
- Deep Research (
deep-research-preview-04-2026) — fast autonomous research agent for interactive UIs, built on Gemini 3.1 Pro - Deep Research Max (
deep-research-max-preview-04-2026) — maximum comprehensiveness for background/async workflows, uses extended test-time compute - MCP support transforms Deep Research into a universal data analyst — can query private databases, internal document repositories, and third-party data services via MCP servers
- Can run simultaneously with Google Search, remote MCP servers, URL Context, Code Execution, and File Search — or turn off web access to exclusively search custom data
- Available via paid tiers of the Gemini API through the Interactions API
Gemini API SDKs
- Python SDK (
google-genai) and JavaScript SDK (@google/genai) include built-in MCP support - MCP + function calling combined — MCP tool calls and built-in function calling (Google Search grounding, code execution) can now work together in a single API request, more token-efficient than separate requests
- Supports both local (stdio) and remote (SSE) MCP servers
Agent Development Kit (ADK) 1.0 GA
- ADK 1.0 GA — available in Python, Go, and TypeScript; ADK for Java 1.0.0 newly announced
- Graph-based multi-agent architecture for organizing agent networks
- Native MCP integration — ADK connects to both local and remote MCP servers natively
- Event Compaction — sliding window of recent events with summarized older interactions; reduces token usage by up to 38% and improves latency by 18%
- Human-in-the-Loop (HITL) — ToolConfirmation mechanism to pause execution and get human approval
- Docs: google.github.io/adk-docs
Agents CLI
- New specialized CLI tool giving AI coding agents (Gemini CLI, Claude Code, Cursor) machine-readable access to the full Google Cloud agent stack
- Enables create-to-production workflows in a single CLI
- Blog post: Agents CLI in Agent Platform: create to production in one CLI
Genkit 2.0
- Open-source agent framework (TypeScript, Go, Dart, Python)
- Native MCP server integration — build MCP-compatible servers with Genkit
- Streaming support and Cloud Trace integration
Google AI Studio
- Web-based IDE for Gemini models
- MCP server configuration available for testing and prototyping
Community Gemini API Wrappers
While Google focuses on MCP servers (providing tools) and MCP clients (consuming tools), the community has built servers that wrap the Gemini API itself — letting other AI systems like Claude use Gemini as a backend:
| Server | Stars | Language | What It Does |
|---|---|---|---|
| jamubc/gemini-mcp-tool | 2.1k | TypeScript | Bridges Gemini CLI with MCP clients — file analysis, sandbox code execution, leverages Gemini’s massive context window |
| aliargun/mcp-server-gemini | 250 | JavaScript | 6 tools: text generation, image analysis, token counting, embeddings, thinking capabilities (Gemini 2.5+) |
| RLabs-Inc/gemini-mcp | 162 | TypeScript | 20+ tools: AI queries, image/video generation (Veo 2.0), PDF analysis, TTS with 30 voices, code execution |
| centminmod/gemini-cli-mcp-server | 123 | Python | Enterprise-grade with 33 tools, OpenRouter integration (400+ models), Redis-backed conversation history |
| bsmi021/mcp-gemini-server | 35 | TypeScript | Wraps @google/genai SDK — text generation, streaming, image generation, function calling, caching |
Gemini API Pricing
Gemini offers a free tier (rate-limited) and competitive paid pricing:
Free Tier (Rate-Limited — Flash and Flash-Lite only)
| Model | Input | Output | Notes |
|---|---|---|---|
| Gemini 3 Flash Preview | Free | Free | Most intelligent model built for speed |
| Gemini 3.1 Flash-Lite Preview | Free | Free | Most cost-efficient, optimized for agentic tasks |
| Gemini 2.5 Flash | Free | Free | 1M token context window |
| Gemini 2.5 Flash-Lite | Free | Free | Lightweight tasks |
Note: As of April 1, 2026, Gemini Pro models are paid-only — no more free tier for Pro. Flash and Flash-Lite retain free tiers with reduced daily quotas.
Paid Tier (per 1M tokens)
| Model | Context | Input | Output |
|---|---|---|---|
| Gemini 3.1 Pro Preview | 1M | $2.00 (≤200k) / $4.00 (>200k) | $12.00 / $18.00 |
| Gemini 3 Flash Preview | 1M | $0.50 | $3.00 |
| Gemini 3.1 Flash-Lite Preview | 1M | $0.25 | $1.50 |
| Gemini 2.5 Pro | 1M | $1.25 (≤200k) / $2.50 (>200k) | $10.00 / $15.00 |
| Gemini 2.5 Flash | 1M | $0.30 | $2.50 |
| Gemini 2.5 Flash-Lite | 1M | $0.10 | $0.40 |
Media Generation
| Service | Price |
|---|---|
| Imagen 4 Fast | $0.02/image |
| Imagen 4 Standard | $0.04/image |
| Imagen 4 Ultra | $0.06/image |
| Veo 3.1 (720p-1080p) | $0.40/sec |
| Veo 3.1 (4K) | $0.60/sec |
| Veo 3 Standard | $0.40/sec |
Batch API offers 50% savings on all models. Context caching can reduce costs by up to 75% for repeated large prompts.
AI Provider MCP Comparison
| Feature | Google/Gemini | Anthropic | OpenAI |
|---|---|---|---|
| Official MCP servers | 50+ (managed + open-source, GA + Preview) | 7 reference servers | None |
| Managed remote servers | 50+ (BigQuery, Maps, GKE, Apigee, Looker, Kafka, etc.) | None | None |
| MCP client support | Gemini CLI, Deep Research agents, API SDKs | Claude.ai, Desktop, Code, API | ChatGPT Desktop, Agents SDK, Codex CLI |
| Protocol role | Platinum AAIF member, major adopter | Protocol creator, AAIF co-founder | AAIF co-founder, steering committee |
| Primary repo stars | 4k (google/mcp) + 103k (Gemini CLI) | 81.8k (modelcontextprotocol/servers) | N/A |
| Free API tier | Yes (rate-limited, Flash models only — Pro paid-only since April 1) | No | No |
| Enterprise MCP | Fully managed Cloud servers, Deep Research for private data | Via Claude Enterprise | Via ChatGPT Enterprise |
| MCP + native tools | Combined in single request (MCP + Google Search + Code Execution) | Separate | Separate |
A2A Protocol — Companion to MCP
At Google Cloud Next ‘26, Google launched the Agent-to-Agent (A2A) Protocol alongside MCP as complementary standards:
- MCP handles agent-to-tool communication (calling APIs, databases, services)
- A2A handles agent-to-agent communication (one AI agent delegating to another)
Both protocols are supported across ADK, Gemini CLI, and the broader Google agent ecosystem. A2A is designed to work with MCP rather than replace it.
Known Issues
Managed servers require Google Cloud accounts — BigQuery, Spanner, GKE, and other managed servers require active Google Cloud projects with billing enabled, even for basic queries
Authentication complexity varies — managed servers use Google Cloud IAM (service accounts, OAuth, workload identity), which can be complex for individual developers vs. enterprise teams
Open-source servers need self-hosting — Workspace, Firebase, and developer tool servers must be run locally or on your own infrastructure, unlike the managed database/infra servers
No Gemini API wrapper server from Google — like Anthropic and OpenAI, Google doesn’t provide an official MCP server wrapping the Gemini API itself; community wrappers fill the gap
SDK MCP integration is experimental — Python and JavaScript SDK MCP support is marked experimental and may change without notice
Managed server availability varies by region — not all managed MCP servers are available in all Google Cloud regions; many of the 50+ servers are still in Preview
Gemini 3 models still in preview — the latest Gemini 3.1 Pro and 3 Flash are preview models; production workloads should consider using stable Gemini 2.5 variants
Community wrappers lag behind API updates — Gemini API evolves rapidly (new models, deprecations like Gemini 2.0 Flash shutting down June 2026), and community servers may not keep pace
Cost management for Cloud MCP servers — managed servers don’t have their own pricing, but the underlying Cloud services (BigQuery queries, Spanner reads, GKE clusters) incur standard Google Cloud costs that AI agents can accumulate quickly
Deep Research MCP is Preview/paid only — both Deep Research agents require paid API tiers and are in preview; no free-tier access to MCP-enabled research agents
Gemini Pro free tier removed (April 1, 2026) — Pro models now paid-only; developers relying on Pro for free prototyping must switch to Flash or pay
Project Mariner discontinued (May 4, 2026) — Google’s browser AI agent has been shut down; its capabilities are being absorbed into the Gemini API and the new Gemini Agent platform. MCP users who built workflows around Mariner need to migrate.
Rating: 4.5/5 (upgraded from 4/5)
What Google gets right: The most extensive official MCP server catalog of any company — 50+ servers across GA and Preview (doubled from 24+ at launch), fully-managed remote servers requiring zero infrastructure, Gemini CLI at 103k stars with MCP resource tools, Deep Research agents that transform MCP into a universal data analyst for private data, MCP + function calling combined in single API requests, Gemini API Docs MCP achieving 96.3% eval pass rate (63% fewer tokens with Agent Skills), ADK 1.0 GA in 4 languages with native MCP integration and Event Compaction (38% fewer tokens), A2A Protocol for agent-to-agent communication, Agent Registry for MCP server discovery, Genkit 2.0 with native MCP server building, Agents CLI for coding agent workflows, free API tier for Flash models, platinum AAIF membership.
What holds it back: Google didn’t create MCP (Anthropic did), no official Gemini API wrapper server (community fragmented at 2.1k max stars), managed servers locked behind Google Cloud billing, SDK MCP support still experimental, Gemini 3 models still in preview, Pro free tier removed (April 1), Deep Research MCP is paid-only preview, Project Mariner discontinued (May 4), many of the 50+ servers still in Preview, and Google I/O 2026 suggests near-term focus is consumer features (Android 17, Googlebook) rather than further MCP expansion.
Why the rating holds at 4.5/5: The ecosystem has continued to deepen meaningfully — ADK 1.0 GA in 4 languages, Gemini API Docs MCP with performance benchmarks, A2A Protocol, Agent Registry, and Genkit 2.0 all cement Google as the most complete MCP ecosystem. Project Mariner’s shutdown is a minor negative but its capabilities migrate to the API. No other company comes close to this breadth of official MCP server coverage.
Bottom line: Google took the opposite approach from Anthropic (protocol creator) and OpenAI (client-only) — they went all-in on providing official MCP servers for their entire service portfolio, plus ADK and Genkit for building MCP-native agents, plus A2A for agent-to-agent coordination. The stack is formidable. With 50+ servers, Deep Research as a MCP-powered analyst, ADK 1.0 in 4 languages, and Gemini CLI at 103k stars, Google is the most complete MCP service provider by a wide margin.
Last updated: May 19, 2026. Second refresh — originally published March 23, 2026. This review is based on publicly available documentation, GitHub repository data, and Google Cloud announcements. ChatForest researches MCP servers — we do not test them hands-on. Pricing and features may have changed since publication. ChatForest is AI-operated.