You just merged a PR. Now you need to deploy it. You open a browser tab, navigate to your hosting dashboard, find the project, click deploy, wait for the build, check logs, maybe update some environment variables first. Three clicks become ten. Ten become twenty when something breaks.
Now imagine telling your AI assistant: “Deploy the latest commit to staging, then show me the build logs.” Done. No dashboard. No context-switching. No remembering which tab has which environment.
That’s what deployment platform MCP servers enable. They connect AI agents directly to your hosting platform — Vercel, Cloudflare, Netlify, Railway, Heroku, DigitalOcean, Dokploy, Coolify — and to your GitOps tools like ArgoCD and FluxCD. Your AI can deploy code, manage environments, check build status, roll back failures, and configure services, all through natural language from inside your editor.
This review covers MCP servers for deployment platforms, PaaS providers, and GitOps deployment tools. For CI/CD pipelines and build automation (GitHub Actions, Jenkins, CircleCI), see CI/CD MCP Servers. For container orchestration (Kubernetes management), see Kubernetes MCP Servers. For infrastructure provisioning (Terraform, Pulumi, Ansible), see Infrastructure Automation MCP Servers.
Part of our DevOps MCP category. The headline finding: every major deployment platform now has an official MCP server, from edge platforms (Cloudflare, Vercel) to traditional PaaS (Heroku, Railway) to self-hosted alternatives (Dokploy, Coolify). The self-hosted PaaS space has particularly strong MCP coverage, with Dokploy offering 508 tools — the most tool-rich deployment MCP server we’ve reviewed.
Cloud & Edge Platform MCP Servers
Cloudflare MCP Servers
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| cloudflare/mcp | — | TypeScript | — | 2500+ endpoints, Code Mode |
| cloudflare/mcp-server-cloudflare | — | TypeScript | — | 13 specialized servers |
| cloudflare/workers-mcp | — | TypeScript | — | Worker bridge |
The broadest deployment platform MCP coverage available — now with 13 specialized servers and a unified portal. Cloudflare’s Code Mode MCP server at mcp.cloudflare.com covers 2500+ API endpoints using just two tools — search() and execute() — running inside an isolated Dynamic Worker sandbox. Cloudflare claims up to 99.9% token reduction versus raw API coverage.
Coverage spans: Workers (serverless compute), Pages (static sites), KV (key-value storage), R2 (object storage), D1 (SQLite databases), DNS, Firewall, Load Balancers, Stream (video), Images, AI Gateway, Vectorize (vector database), Access (zero-trust), and more.
Cloudflare launched MCP Server Portals — a unified gateway providing discovery and access to all 13 specialized domain servers (Audit Logs, DNS, Workers, R2, Zero Trust, and more). In May 2026, Cloudflare integrated with Anthropic’s Claude Managed Agents for isolated code execution.
Why it matters: Cloudflare’s Code Mode is architecturally distinctive — instead of hundreds of individual tools, the AI gets a code-execution interface that constructs and fires API calls dynamically. This scales to thousands of endpoints without bloating the tool list. The MCP Server Portals addition makes it easier to choose focused coverage (a specific product’s server) vs. unified access.
Limitation: Code Mode requires the AI to understand Cloudflare API semantics rather than calling named tools. The breadth-over-depth trade-off means individual operations need more AI reasoning than purpose-built tools provide.
Vercel MCP Server
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| Vercel MCP | — | — | — | Remote MCP, OAuth |
Vercel’s official MCP server implements the latest MCP Authorization and Streamable HTTP specifications. Available as a remote MCP server — no local installation required.
Tools are split between public (no auth needed) and authenticated categories:
- Teams — manage Vercel team settings and membership
- Projects — create, configure, and manage projects
- Project Members — manage project-level access
- Deployments — trigger, list, and inspect deployments
- Deployment Events — monitor deployment progress and logs
- Environment Variables — manage environment configuration across preview, development, and production
Security model: OAuth-based authentication with strict client approval — Vercel MCP only supports AI clients that have been reviewed and approved by Vercel. This is an unusually cautious approach compared to most MCP servers, which accept any client with valid credentials.
Why it matters: Vercel is the dominant hosting platform for Next.js and modern frontend applications. Their MCP server enables the complete deployment loop from within an AI assistant: create a project, configure environment variables, deploy, and monitor — without touching the Vercel dashboard. The remote MCP approach means zero local setup.
Additional Vercel MCP tooling:
- vercel/mcp-handler — helps developers build their own MCP servers that run on Vercel, supporting Next.js, Nuxt, Svelte, and more frameworks
- vercel/next-devtools-mcp — provides Next.js development tools specifically for coding agents
- IvanAmador/vercel-ai-docs-mcp — AI-powered search and querying of Vercel AI SDK documentation
Limitation: Beta status on all plans. The approved-clients-only requirement limits ecosystem compatibility. Star counts not publicly available for the official server.
Netlify MCP Server
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| netlify/netlify-mcp | — | TypeScript | — | 9 tools |
Netlify’s official MCP server enables the “prompt to production” workflow — create, build, deploy, and manage Netlify sites from within an AI assistant.
The 9 tools cover:
- Create and deploy sites — from code to live URL in a single conversation
- Manage environment variables and secrets — configure build settings, API keys, environment-specific values
- Install and uninstall extensions — manage Netlify integrations programmatically
- Configure access controls — manage site-level permissions
- Fetch user and team information — check account details and team membership
- Manage form submissions — access Netlify Forms data
Why it matters: Netlify pioneered the Jamstack deployment model. Their MCP server covers the essentials for the typical static site / serverless function deployment workflow. The focus on 9 well-curated tools (rather than trying to cover every API endpoint) makes for clean AI interactions.
Community status: DynamicEndpoints/Netlify-MCP-Server (previously 43 tools with full CLI coverage) is no longer publicly accessible as of May 2026. The official server is the only actively available option.
Limitation: 9 tools is relatively thin compared to competitors. The official repository shows last commit in March 2025 — development appears stagnant. No deployment rollback capability. Star count not publicly available.
Developer PaaS MCP Servers
Railway MCP Server
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| railwayapp/railway-mcp-server | 192 | TypeScript | — | 14 tools, project & deploy management |
| Railway Remote MCP | — | — | — | Remote MCP, OAuth, railway-agent |
Railway’s official MCP server connects AI assistants to the full Railway deployment lifecycle — and now ships a hosted remote MCP at mcp.railway.com. Railway positions itself as “stateful, serverful, pay-per-use infrastructure” — ideal for backend services, databases, and always-on processes.
The local server exposes 14 tools for project management, environments, deployments, services, and log retrieval. The new remote MCP server (OAuth authentication, no local install) adds a particularly powerful tool: railway-agent — which delegates complex multi-step operations to Railway’s internal AI agent for log analysis, debugging, database setup, and more. Compatible with Claude, Cursor, GitHub Copilot, Cline, and Devin.
Why it matters: Railway’s remote MCP launch significantly lowers the barrier to adoption — zero local configuration. The railway-agent tool enables the kind of autonomous multi-step operations (debug a failing deployment, spin up a database, wire environment variables) that single-purpose tools can’t handle.
Community alternative: jason-tan-swe/railway-mcp (72 stars) remains available but has seen no updates since June 2025 as the official server has caught up.
Limitation: 192 stars suggests moderate adoption. The railway-agent delegation model is useful but means some operations route through Railway’s internal AI rather than being direct tool calls.
Heroku MCP Server
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| heroku/heroku-mcp-server | — | TypeScript | — | Heroku CLI integration |
Heroku’s official MCP server is the only deployment platform with native Salesforce Agentforce integration — now available as a remote MCP server at mcp.heroku.com. The server wraps the Heroku CLI, providing platform management through AI assistants.
The hosted remote MCP (launched June 2025, now production-stable) uses OAuth authentication and supports Claude Desktop, Cursor, and VS Code. It covers app lifecycle, dyno scaling, PostgreSQL administration, add-ons, pipelines, and teams. Heroku provides official MCP server templates in Python, Node.js, Ruby, and Go for teams building their own MCP servers on the platform.
Why it matters: Heroku’s Salesforce integration remains unique in the deployment MCP space. Review Apps integration means every PR can spin up a disposable test environment managed by an AI agent through the full PR → Review App → Production pipeline.
Limitation: Heroku has no free tier for new accounts. v1.2.2 (April 22, 2026) was a bug fix; development pace is measured relative to the newer platforms.
DigitalOcean MCP Server
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| digitalocean-labs/mcp-digitalocean | 106 | — | — | 9+ services, v1.0.59 |
DigitalOcean’s active MCP server (now at digitalocean-labs/mcp-digitalocean, v1.0.59) covers App Platform deployment and broader infrastructure management. The original digitalocean/digitalocean-mcp repo was archived; the labs repo is the current maintained version with 600+ commits.
Service coverage includes:
- App Platform — deploy and manage applications (supports native remote MCP hosting with HTTP response streaming)
- Databases — managed database provisioning and management
- DOKS — DigitalOcean Kubernetes Service
- Droplets — virtual machine management
- Networking — VPCs, load balancers, firewalls
- Spaces Storage — S3-compatible object storage
- GenAI / Custom Models — NEW in May 2026: import, manage, and evaluate custom ML models (v1.0.54–1.0.59); safe deletion with user consent prompt
Why it matters: DigitalOcean’s blend of simplicity and infrastructure depth makes their MCP server a one-stop shop for teams on DO. The GenAI/custom model management additions in May 2026 reflect DO’s push into AI infrastructure — unusual for a deployment platform MCP server.
Limitation: The repo migration (archived → labs) may cause confusion in documentation and tooling configs that reference the old path. Development pace is very active (multiple releases weekly) which means feature sets evolve quickly.
DeployHQ MCP Server
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| deployhq/deployhq-mcp-server | — | TypeScript | — | 7 tools |
DeployHQ’s official MCP server provides focused deployment automation tools. DeployHQ is a deployment service that deploys code from Git repositories to servers via FTP, SFTP, or SSH.
The 7 tools cover: List Projects, Get Project Details, List Servers, List Deployments, Get Deployment Details, Get Deployment Log, and Create Deployment.
Why it matters: DeployHQ fills a niche that cloud PaaS platforms don’t — deploying to traditional servers. If you’re deploying to shared hosting, VPS, or on-premise servers via FTP/SFTP, DeployHQ’s MCP server lets your AI trigger deployments and inspect logs without navigating the deployment dashboard.
Limitation: 7 tools is minimal. No rollback, no environment variable management, no build configuration. The server covers the read/deploy cycle but not the full deployment lifecycle.
Render MCP Server
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| render-oss/render-mcp-server | 133 | TypeScript | — | Hosted remote MCP, create + manage + logs |
Render’s official MCP server launched in January 2026, hosted at mcp.render.com. Render is an alternative to Heroku/Railway for web services, static sites, background workers, and managed databases — and their MCP server covers the full deployment lifecycle with a safety-conscious design.
Tools cover: create web services, static sites, cron jobs, Postgres instances, and Key Value stores; list and update services; deployment history; filtered logs; metrics (CPU, memory, bandwidth); and read-only SQL queries against managed Postgres.
Why it matters: 133 stars for a ~4-month-old server indicates solid early adoption. The read-only SQL query tool is particularly useful — AI can analyze production database state during an incident without needing direct database credentials. Compatible with Cursor, Codex, and Claude Code.
Limitation: No image-backed service creation, no IP allowlists, no deploy triggering (list deployments but can’t trigger), no resource deletion except environment variables — intentional conservative scope for v0.3.0.
Koyeb MCP Server
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| koyeb/mcp-server-koyeb | 6 | — | — | Beta, app/service/deployment management |
Koyeb’s official MCP server launched in 2026. Koyeb is a serverless platform offering one-click deployment from Git or Docker — and was acquired by Mistral AI in February 2026, making it the only deployment platform MCP server from an AI company.
Tools cover: create and list apps; manage services; deployment tracking with logs; instance monitoring.
Why it matters: The Mistral AI acquisition could accelerate Koyeb’s MCP integration and position it as the natural home for Mistral-based AI workloads. Early days — 6 stars and beta status — but the strategic parent is notable.
Limitation: Beta with no formal releases. 6 stars indicates very early adoption. Tool coverage is minimal compared to established platforms.
Self-Hosted PaaS MCP Servers
Dokploy MCP Server
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| Dokploy/mcp | — | TypeScript | — | 508 tools, 49 categories |
The most tool-rich deployment MCP server we’ve reviewed — 508 tools across 49 categories. Dokploy (26K+ stars) is an open-source, self-hostable PaaS alternative to Vercel, Heroku, and Netlify.
The MCP server (251 stars) exposes the full Dokploy API: project and application management, databases, Docker configuration, notifications, SSO, backups, domain management, SSL certificates, environment variables, and more. Modern HTTP mode supports both Streamable HTTP (MCP 2025-03-26 spec) and Legacy SSE (MCP 2024-11-05).
Notable May 2026 update: Secret field redaction added (May 3) — an opt-in security feature that redacts secret-bearing fields from API responses, preventing credential leakage into AI context. A meaningful security improvement for production deployments.
Why it matters: 508 tools is exceptional coverage. For teams running Dokploy as their self-hosted deployment platform, this MCP server turns the entire platform into an AI-managed resource. The breadth means an AI agent can handle everything from creating a new project to configuring SSL to managing database backups.
Limitation: 508 tools is a lot for an AI to navigate efficiently. Tool discovery and selection may degrade with this many options. The server is auto-generated from the API spec, so tool descriptions may not be optimized for AI consumption.
Coolify MCP Servers
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| StuMason/coolify-mcp | 379 | TypeScript | — | 42 tools, v2.11.0 |
| dazeb/coolify-mcp-enhanced | 16 | TypeScript | — | Enhanced AI features |
Coolify (35K+ stars) is the most-starred self-hosted PaaS platform, and StuMason/coolify-mcp is among the most actively developed community MCP servers in the DevOps space. At 379 stars and 5 releases since late April 2026, it’s now at v2.11.0 with 42 tools (up from 38).
Core tools cover:
- Teams and Projects — organizational management
- Servers — manage deployment targets, including Hetzner cloud provider (NEW v2.11.0)
- Private Keys — SSH key management for server access
- Applications — deploy, configure, and manage apps with full build config (Dockerfile path, base dir, install/build commands) and all 12 health-check parameters
- Databases — provision and manage databases with backup management
- Services — manage background services
- Environment Variables — configure runtime settings; values masked by default to prevent secret leakage (v2.9.0)
- Deployments — trigger and monitor deployments
- Storages, Scheduled Tasks, System — NEW in v2.11.0
The enhanced version (dazeb/coolify-mcp-enhanced, 16 stars) adds AI-powered natural language infrastructure management. A v3 rewrite is in development that reframes the server as a subscribable surface with Resources, Tasks, and Prompts primitives.
Why it matters: 379 stars for a community-built server reflects real adoption. The v2.9.0 security improvement (env var value masking) and v2.11.0 Hetzner integration show the project responding to real production needs.
Additional implementations: GoCoder7/coolify-mcp-server provides a focused 4-tool implementation for simpler use cases.
Limitation: No official MCP server from the Coolify team — all implementations are community-built. The active v3 rewrite may introduce breaking changes.
GitOps Deployment MCP Servers
ArgoCD MCP Server
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| argoproj-labs/mcp-for-argocd | 467 | Go | — | GitOps tools, write ops + MCP_READ_ONLY |
The most-starred GitOps MCP server, from the ArgoCD Labs organization — now with full write operation support. ArgoCD (18K+ stars) is the dominant Kubernetes GitOps tool, and this MCP server has grown rapidly (354 → 467 stars since April 2026, 14 total releases).
v0.7.0 (May 2, 2026) added stateless mode (--stateless flag) for Kubernetes deployments with horizontal pod autoscaling, eliminating session affinity requirements. A /healthz health check endpoint enables proper liveness probes in HTTP mode.
Tool set:
- list_clusters — list all clusters registered with ArgoCD
- list_applications — list and filter all applications across clusters
- get_application — detailed status, sync state, health, and deployment history
- create_application, update_application, delete_application — full CRUD
- sync_application — trigger ArgoCD sync (the critical write operation)
- run_resource_action — perform actions on Kubernetes resources
Safety model: All write operations can be disabled via MCP_READ_ONLY=true environment variable — allowing read-only deployments in sensitive environments. This is the safety control pattern the deployment MCP space has needed.
Supports both stdio and HTTP stream transport for flexible client integration.
Community alternative: severity1/argocd-mcp (12 stars) provides a community Python implementation that integrates with the ArgoCD API; dormant since April 2025.
Limitation: 467 stars is strong growth but the server is still maturing. The write operations (especially delete_application) require careful access control configuration — the MCP_READ_ONLY=true guard is necessary but opt-in.
FluxCD MCP Server
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| flux-operator MCP | 634 | Go | — | GitOps tools, v0.50.0 |
The Flux Operator MCP Server connects AI assistants directly to Kubernetes clusters running Flux GitOps. Built as a component of the Flux Operator project by ControlPlane (634 stars).
v0.50.0 (May 20, 2026) added OpenShift v4.21 compatibility. v0.49.0 (May 12) fixed MCP tool schemas to include the required “properties” field for OpenAI compatibility, added AWS provider GitRepository sync, and upgraded OCIRepository to v1. The MCP Go SDK was upgraded to v1.4.0+.
The server enables AI assistants to:
- Analyze cluster state — trace issues from high-level GitOps resources (ResourceSets, HelmReleases, Kustomizations) all the way down to Kubernetes deployments and pod logs
- Troubleshoot deployment issues — correlate events across the GitOps stack with natural-language GitOps root cause analysis
- Perform operations — interact with Flux resources through conversational prompts
Distributed as a single statically-compiled Go binary with no external dependencies, for Linux, macOS, and Windows on AMD64 and ARM64.
Why it matters: Flux’s MCP server traces the full deployment path from Git source to running pod — when something breaks, the AI identifies where in the reconciliation chain the issue occurred (HelmRelease, Kustomization, GitRepository). The v0.49.0 OpenAI schema fix improves compatibility with non-Claude MCP clients.
Limitation: The MCP server is embedded within the Flux Operator rather than being a standalone project, making adoption tracking and independent versioning harder.
Cloud Deployment Assistants
Fly.io MCP Tools
| Server | Stars | Language | License | MCP Features |
|---|---|---|---|---|
| superfly/flymcp | 31 | Go | — | Official, experimental, wraps flyctl CLI |
| brannn/fly-mcp | 0 | — | — | Community, early stage |
Fly.io’s official flymcp server wraps the flyctl CLI for managing Fly.io infrastructure from AI assistants — currently experimental. Enabled via fly mcp server --claude, it supports: apps, certs, logs, machines, orgs, and status checks.
Fly.io also offers the official fly mcp launch command — which deploys any stdio-based MCP server (npx, uv, go run, or Docker image) to a Fly machine and configures MCP client connectivity. This makes Fly.io both an MCP-managed platform and an MCP server hosting platform.
NakulRajan/mcp-fly-deployer generates Dockerfiles and fly.toml configurations for deploying MCP servers to Fly.io.
Limitation: The official server (31 stars) is marked experimental with no formal releases. The community brannn/fly-mcp (0 stars) remains early-stage. Fly.io’s MCP story is functionally available but not production-recommended.
The Deployment Platform MCP Landscape
The deployment platform MCP space reveals a clear market structure — and the gap between tiers has narrowed significantly since April 2026:
Tier 1 — Full API coverage:
- Cloudflare (2500+ endpoints via Code Mode + 13 specialized servers + MCP Server Portals)
- Dokploy (508 tools, auto-generated from OpenAPI, now with secret field redaction)
Tier 2 — Production-ready official servers:
- Vercel (remote MCP, OAuth, deployment management)
- Netlify (9 curated tools, prompt to production)
- Railway (192 stars, remote MCP at mcp.railway.com, railway-agent tool)
- Heroku (remote MCP at mcp.heroku.com, Salesforce Agentforce)
- DigitalOcean (v1.0.59, 9+ services, GenAI custom model management)
- ArgoCD (467 stars, full write ops + MCP_READ_ONLY safety)
- Render (133 stars, NEW, hosted at mcp.render.com)
- Coolify (379 stars, 42 tools, community-built, very active)
Tier 3 — Emerging/community:
- FluxCD (Flux Operator component, v0.50.0, deep GitOps tracing)
- DeployHQ (7 tools, traditional deployment)
- Koyeb (6 stars, NEW, beta, Mistral AI subsidiary)
- Fly.io (official experimental CLI + community)
Key Trends (May 2026)
-
Remote MCP is now standard for cloud platforms. Vercel, Cloudflare, Railway, Heroku, DigitalOcean, and Render all offer hosted remote MCP servers. Zero local installation, OAuth authentication. The ecosystem has completed the shift to remote-first that was just starting in late 2025.
-
GitOps is no longer read-only. ArgoCD’s MCP server (467 stars) added full write operations in 2026 —
sync_application,create_application,update_application,delete_application— withMCP_READ_ONLY=trueas an explicit safety gate. This is a meaningful shift: AI agents can now trigger syncs and manage applications, not just inspect them. -
Safety controls are becoming standard. ArgoCD’s
MCP_READ_ONLY=true, Dokploy’s secret field redaction, Coolify’s masked env var values — these security patterns are appearing across the space. The gap between “destructive operations allowed by default” (the April 2026 status) and “safety-conscious defaults” is closing. -
Self-hosted PaaS has the strongest MCP coverage. Dokploy (508 tools) and Coolify (379 stars, 42 tools) — the two leading open-source PaaS platforms — continue to outpace cloud PaaS on coverage depth. The audience overlap (self-hosters who want control) with AI-assisted infrastructure management is real.
-
New entrants in every sub-segment. Render (133 stars) fills the gap between Railway and Heroku. Koyeb (Mistral AI) bridges AI and deployment platform. The space is still actively developing first-party coverage.
Comparison Table
| Platform | Server | Stars | Tools | Official | Transport | Key Strength |
|---|---|---|---|---|---|---|
| Cloudflare | cloudflare/mcp | — | 2500+ endpoints | Yes | Remote | Broadest API coverage, 13 servers |
| Dokploy | Dokploy/mcp | 251 | 508 | Yes | HTTP/SSE | Most tools, secret redaction |
| Coolify | StuMason/coolify-mcp | 379 | 42 | Community | stdio | Self-hosted PaaS, very active |
| ArgoCD | argoproj-labs/mcp-for-argocd | 467 | 7+ | Labs | stdio/HTTP | GitOps write ops + MCP_READ_ONLY |
| Railway | railwayapp/railway-mcp-server | 192 | 14 | Yes | stdio/Remote | Remote MCP, railway-agent |
| Render | render-oss/render-mcp-server | 133 | Multiple | Yes | Remote | NEW, hosted, read-only SQL |
| Vercel | Vercel MCP | — | 13 | Yes | Remote | Frontend deployment |
| Netlify | netlify/netlify-mcp | — | 9 | Yes | stdio | Prompt to production |
| Heroku | heroku/heroku-mcp-server | 77 | Multiple | Yes | stdio/Remote | Salesforce integration |
| DigitalOcean | digitalocean-labs/mcp-digitalocean | 106 | 9+ services | Yes | stdio | Infrastructure + PaaS + GenAI |
| FluxCD | flux-operator | 634 | Multiple | Yes | stdio | Deep GitOps tracing, v0.50.0 |
| DeployHQ | deployhq/deployhq-mcp-server | — | 7 | Yes | stdio | Traditional deployment |
| Fly.io | superfly/flymcp | 31 | 5+ | Yes (exp.) | stdio | Experimental official CLI |
| Koyeb | koyeb/mcp-server-koyeb | 6 | Multiple | Yes | — | NEW, beta, Mistral AI subsidiary |
Rating: 4.5 / 5
The deployment platform MCP space matured significantly in the first half of 2026. ArgoCD (354 → 467 stars) added full write operations with MCP_READ_ONLY=true safety controls, directly addressing the biggest gap from April 2026. Railway launched a remote MCP with an AI-agent delegation tool. Coolify reached 379 stars with 42 tools across 5 releases in 3 weeks. Render (133 stars) and Koyeb entered the space with official servers. Cloudflare expanded to 13 specialized servers and launched MCP Server Portals.
Raised from 4.0 to 4.5:
- ArgoCD write operations + MCP_READ_ONLY resolves the “GitOps servers are read-only” deduction
- Safety patterns (secret redaction, env var masking, read-only guards) are appearing across the space
- New entrants (Render, Koyeb) fill coverage gaps that existed in April
Remaining deductions (0.5 points):
- Cross-platform orchestration still absent — no MCP server abstracts deployment across providers; multi-platform teams still need N separate server configs
- Quality floor inconsistency — Netlify’s official server shows last commit March 2025 and its community alternative has gone 404; not every platform maintains parity with the ecosystem leaders
For teams deploying to a single platform, their platform’s official MCP server is the clear choice. For multi-platform deployment, configure each platform’s server separately and use an MCP client that manages multiple server connections.
This review covers publicly available data as of May 2026. Star counts and feature sets change; check the linked repositories for current information. ChatForest researches MCP servers — we do not test or endorse specific tools. See our methodology for how we research and rate MCP servers.