Best Infrastructure Plugins, Skills, and MCP Servers in 2026
Research-backed infrastructure plugins, skills, and MCP servers with platform availability and direct setup routes.
Updated July 25, 2026
These extensions connect your agent to cloud platforms, deployments, infrastructure-as-code, container and cluster operations, and production observability. The central trade-off is access: read-only guidance changes nothing, while live credentials let an agent alter production systems. Grant the narrowest permission the task needs.
A cross-host package that combines Azure MCP, Foundry MCP, and curated Azure skills. It pairs current decision guidance with more than 200 structured tools for inspecting resources, validating deployments, diagnosing infrastructure, and optimizing costs.
Reach for it when building, deploying, or operating Azure and Microsoft Foundry workloads and you want the agent grounded in current practice rather than guessing. Give it least-privileged identities, and note that sovereign clouds need explicit MCP configuration.
A broad bundle spanning Vercel, Next.js, React, the AI SDK, deployments, performance, and infrastructure. It ships product-specific skills, specialist agents, operational commands, project-aware hooks, an ecosystem map, and, on supported routes, live Vercel account access.
Its value is highest in projects built on Vercel or its ecosystem. The skills help without any login, but deployments, logs, and environment changes need Vercel authentication. In unrelated projects the broad context is mostly noise, so a narrow route serves better.
A developer package that combines Cloudflare’s API MCP with skills for Workers, Durable Objects, the Agents SDK, Wrangler, MCP servers, sandboxing, and web performance. The agent gets both current platform patterns and the ability to inspect or change live Cloudflare resources.
Use it when building or operating applications on Cloudflare, from Workers and storage to databases, deployments, logs, and analytics. Be deliberate about tokens: OAuth or bearer-token access can permit production writes, including deployments and security changes, so scope it to what the task requires.
Docker’s MCP Catalog and Toolkit: a gateway that runs MCP servers as isolated containers, a curated catalog of containerized servers, and Docker Desktop integration for connecting them to agent clients with unified configuration and secrets.
Use it when you already run Docker and want one controlled local layer for many MCP servers rather than configuring each separately. It is not the simplest path to a single hosted connector, and every enabled server still carries its own permissions and supply-chain risk.
An agent stack that combines Firebase skills, a local MCP server built into firebase-tools, and the Firebase CLI. It gives the agent current guidance plus project-aware access to Firestore, Authentication, Functions, Hosting, Storage, Crashlytics, Messaging, Remote Config, and security rules.
Use it to build, configure, diagnose, and deploy Firebase apps with live project context instead of generic advice. Because it can change real resources, review security rules and deployments before they apply, and expose only the feature groups a task actually needs.
A managed MCP paired with a family of installable plugins: a general AWS Core plus specialist packages for agents, analytics, and DevSecOps. It gives agents current AWS guidance and auditable access to AWS APIs, documentation, and sandboxed execution.
Start with AWS Core for general work and add only the specialist package a project needs rather than loading the whole collection. Live operations require AWS credentials, so use least-privileged, agent-aware IAM policies. AWS now recommends this toolkit as the successor to its Labs MCP servers.
A hosted MCP for production issues and traces, packaged with host-specific plugins for debugging, code review, and instrumentation. It lets an agent investigate real errors, inspect events and stack traces, quantify user impact, and connect findings back to the code.
Use it for error triage, production debugging, and adding Sentry instrumentation, so the agent works from real events rather than symptoms alone. Hosted access uses OAuth and can be constrained to an organization or project; the self-hosted server needs explicit token scopes.
A server that speaks directly to the Kubernetes API - pods, deployments, services, logs, events, and apply operations - through typed tools rather than parsed kubectl output. It ships read-only and destructive-operation controls, and is the most active and widely adopted of the community Kubernetes routes.
Reach for it to troubleshoot and operate clusters with the agent working from real resource state instead of shell string-parsing. Cluster credentials are high impact and write operations hit live resources, so scope the kubeconfig context and prefer a read-only mode for exploration.
The surface centers on HashiCorp’s MCP server, which grounds the agent in current registry, module, provider, policy, and HCP Terraform context. Alongside it, portable skills cover Terraform style, testing, refactoring, stacks, and provider development.
Together the server and skills cut stale syntax, invented provider arguments, and poorly reviewed infrastructure changes. Infrastructure credentials are high impact, so prefer read-only tokens and narrow toolsets, and keep plan and apply authority separate from plain registry or documentation lookups.
A surface that spans platform skills, MCP account tools, and the Netlify CLI. The skills carry current guidance for Functions, Edge Functions, Blobs, Netlify DB, Forms, CDN, frameworks, Identity, and AI Gateway; authorized routes can create sites, deploy code, and manage variables.
Use it for projects hosted on Netlify or when building against its primitives. Knowledge-only skills need no account access, but deployments and account changes do. Review any action that modifies secrets, access controls, production deployments, or form data before it runs.
An MCP integration for dashboards, metrics, logs, alerts, incidents, and related observability data. It comes two ways: a hosted Grafana Cloud endpoint and an open-source server you run yourself. Either gives the agent structured tools instead of scraped dashboards or hand-built API calls.
Pick the hosted route for the simplest Grafana Cloud setup, or the open-source server for self-managed Grafana and tighter control. Permissions flow through Grafana’s user or service-account RBAC, so use read-only mode and the smallest practical tool set on production systems.
Official routes that let an agent query production logs, metrics, traces, dashboards, monitors, incidents, and CI health. It grounds debugging in live telemetry, so the agent can correlate symptoms across services, summarize likely causes, and draft or take selected actions.
Reach for it during production debugging, alert tuning, service-health checks, and evidence-backed incident work. It can see sensitive production telemetry and may support writes, so confirm consequential actions. Similarly named community Datadog servers are separate products and do not share this official route’s status.
An operational skill paired with the Railway CLI and local or hosted MCP access. It gives the agent structured ways to create projects, deploy code, provision databases and storage, manage environments and networking, inspect logs and metrics, and troubleshoot failed services.
The skill also teaches the agent when to use the CLI, local MCP, hosted MCP, or Railway’s API. Local MCP fits when a logged-in machine and current project state matter; hosted MCP fits browser OAuth without a local CLI. Treat deployment, variable, and production-log access as sensitive.
PagerDuty’s official integration connects an agent to your incident-response platform - incidents, on-call schedules, services, and escalation policies - through a hosted MCP server, packaged as a plugin for Claude and Cursor and available as a generic MCP route elsewhere.
During an incident the agent can pull who is on call, what changed, and which services are affected without leaving the terminal; outside incidents it turns schedule and service questions into one-line queries. Write operations reach real incidents and escalation policies, so scope API access deliberately.
Pulumi’s official agent surface: a hosted MCP server for live stack, resource, registry, and policy context, paired with a collection of published skills for authoring, previewing, and deploying infrastructure-as-code programs in general-purpose languages.
Use the MCP for live Pulumi Cloud work and the skills when the need is authoring guidance rather than cloud access. Previews are safe to run freely, but deployments follow your cloud credentials, so gate real deploys behind review and keep token and organization scope tight.
