Infrastructure as Code changed how teams manage cloud resources — Terraform configurations, Ansible playbooks, Pulumi programs, OpenTofu modules. Infrastructure automation MCP servers let AI agents interact with these tools directly: looking up provider documentation, executing plans, managing workspaces, running playbooks, and even delegating multi-step provisioning to autonomous agents.
The headline finding: IaC MCP is rapidly expanding beyond documentation into execution and governance. HashiCorp’s Terraform MCP server (~1,500 stars) has moved well past the v0.5.2 release covered in earlier passes of this review, reaching v1.2.0 (August 2026) via 1.0.0 (June 2026) and 1.1.0 (July 2026) — building on the plan/apply tools, Stacks support, and policy set management added earlier in 2026. Pulumi continues with Neo delegation for autonomous provisioning, though the open-source pulumi/mcp-server GitHub repository is gone — Pulumi has consolidated to a remote-only hosted MCP server (see the Pulumi section below). Red Hat open-sourced ansible/aap-mcp-server — a public GitHub repo with full multi-service AAP coverage — but per Red Hat’s own AAP 2.7 announcement (June 2026), the server remains a Technology Preview, not GA (see the Ansible section for the correction). Two additional gaps from our April review are now filled: Spacelift launched spacelift-io/spacelift-intent (137 stars, natural language provisioning via provider APIs) and Terramate shipped an MCP server (5 stars) with dedicated drift detection. The remaining gap is configuration management — Chef (EOL November 2026), Puppet/OpenVox, and Salt still have zero MCP presence. Part of our Cloud & Infrastructure MCP category.
The Landscape
Terraform / HCP Terraform
| Server | Stars | Language | Tools | Transport |
|---|---|---|---|---|
| hashicorp/terraform-mcp-server | ~1,500 | Go | 40+ | stdio, StreamableHTTP |
| severity1/terraform-cloud-mcp | ~23 | Python | 60+ | stdio |
| nwiizo/tfmcp | ~371 | Rust | 34 | stdio |
| CodeSSRockMan/terraform-plan-mcp-server | — | — | ~6 | stdio |
HashiCorp’s official terraform-mcp-server is the ecosystem leader. ~1,500 stars, ~192 forks, 411+ commits (verified against the live repo — up from the 1,400 stars/146 forks/354 commits recorded in the prior audit). The server crossed a significant architectural threshold in v0.5.x — it now includes plan and apply tools alongside its registry intelligence — and has kept shipping since: v1.0.0 (June 9, 2026), v1.1.0 (July 14, 2026), and the current v1.2.0 (August 4, 2026). This is a major shift from the original documentation-only philosophy.
v0.4.0 (January 2026) added Terraform Stacks support (list_stacks, get_stack_details), policy set management (list_workspace_policy_sets, attach_policy_set_to_workspaces), get_token_permissions, and granular --toolsets/--tools flags for selective capability enablement. v0.5.0 (April 2026) added five plan/apply tools for structured JSON plan output and execution logs, bearer token authorization, heartbeat interval configuration for load-balanced environments, and OpenTelemetry instrumentation for tool usage metrics. v0.5.1 fixed TLS flag propagation for air-gapped environments.
Correction (2026-08-15 audit): a prior version of this review stated v0.5.2 (April 28, 2026) was “the current release.” HashiCorp has since shipped v1.0.0, v1.1.0, and v1.2.0 — see the full release history for changes past v0.5.2.
40+ tools now organized into toolsets:
| Toolset | Key Tools |
|---|---|
| Registry | search_providers, get_provider_details, search_modules, get_module_details, search_policies |
| Registry-Private | Private registry access for enterprise providers/modules |
| Terraform | Workspace management, variables, runs, plan/apply operations, Stacks, policy sets, token permissions |
Install: go install github.com/hashicorp/terraform-mcp-server/cmd/terraform-mcp-server@latest or Docker hashicorp/terraform-mcp-server:1.2.0 (tag tracks the current release). Dual transport: stdio for local dev, StreamableHTTP at localhost:8080/mcp for remote setups. Tool filtering via --toolsets (registry, registry-private, terraform) and --tools flags.
severity1/terraform-cloud-mcp (23 stars, 80 commits, Python, v0.8.20) fills the gap HashiCorp deliberately left: full Terraform Cloud API coverage. 60+ tools spanning workspaces, runs, plans, applies, projects, organizations, state versions, variable sets, cost estimation, and assessment results. Features safe_delete_workspace() with explicit ENABLE_DELETE_TOOLS=true guard for destructive operations and audit-safe response filtering. If you manage HCP Terraform at scale, this is the operational complement to HashiCorp’s server.
nwiizo/tfmcp (371 stars, Rust, MIT, v0.2.2) takes the opposite philosophy: full execution. Init, plan, apply, destroy, state management, workspace operations, import, taint/untaint, formatting, and dependency graphing. Plus security scanning with secret detection, risk-scored plan analysis, drift detection, and module health metrics. v0.2.0 (March 26, 2026) expanded to 34 tools — adding policy search and provider capability search tools, toolset/individual tool filtering flags (mirrors the --toolsets/--tools pattern HashiCorp adopted), and upgraded the MCP SDK from rmcp 0.12 to 1.2. The project has since shipped v0.2.2 (August 13, 2026), updating the MCP SDK further to RMCP 3.0.1 and aligning with the 2026-07-28 MCP protocol revision. Available via Homebrew (brew install tfmcp). Built on Rust Edition 2024 (requires Rust 1.85.0+). This is the “give my AI agent full Terraform CLI access” option — powerful but risky in production.
CodeSSRockMan/terraform-plan-mcp-server adds plan analysis with Webex Teams integration for automated workflow notifications. Niche but useful for CI/CD pipeline integration.
Ansible
| Server | Stars | Language | Tools | Transport |
|---|---|---|---|---|
| ansible/aap-mcp-server | ~32 | TypeScript | 20+ | stdio |
| Ansible Development Tools MCP | — | Python | 10+ | stdio |
| ansible-collections/ansible.mcp | ~10 | Python | MCP plugins | stdio |
| bsahane/mcp-ansible | ~30 | Python | 35+ | stdio |
| sibilleb/AAP-Enterprise-MCP-Server | ~31 | Python | 50+ | stdio |
| tarnover/mcp-sysoperator | ~26 | TypeScript | 15+ | stdio (archived) |
Ansible has the broadest MCP server ecosystem of any IaC tool — six distinct options spanning official, semi-official, and community implementations. Red Hat has also expanded the ecosystem with the ansible.mcp collection enabling playbooks to dynamically discover and call MCP servers.
ansible/aap-mcp-server (32 stars, TypeScript, Node.js 22+) is the official open-source Red Hat AAP MCP server. Red Hat open-sourced the server at github.com/ansible/aap-mcp-server, and it’s publicly installable today, but it is not GA — Red Hat’s own June 2026 “What’s new in AAP 2.7” developer blog post still describes the “MCP server-enabled assistant” as Technology Preview, consistent with the original Red Hat blog announcement that introduced it as a technology preview feature bundled with Ansible Automation Platform 2.6.4. Neither the GitHub README nor the repo’s About section makes a GA claim either.
Correction (2026-08-15 audit): a prior version of this review stated the server had “reached GA” / was “now at general availability.” That was unverifiable against Red Hat’s own sources and is corrected above — as of the most recent primary source (June 2026), it remains a Technology Preview.
Key features: multi-service support covering EDA (Event-Driven Ansible), Controller, Gateway, and Galaxy; YAML-based configuration with RBAC toolsets for fine-grained access control; Prometheus metrics for observability. Two modes: read-only for safe querying and monitoring, or read-write for AI agents to execute jobs and implement changes. Red Hat positions Ansible as the “execution layer for agentic AI.”
Ansible Development Tools MCP (tech preview, documented March 2026) is a separate offering focused on development workflows rather than platform operations. Provides access to the Zen of Ansible design philosophy, best practices, virtual environment management, project scaffolding (playbooks, collections), Ansible Lint with auto-fixing, Execution Environment Builder, and Ansible Navigator for playbook execution. Requires Python 3.11+ and Node.js 18+. Communicates via JSON-RPC 2.0 over stdio.
ansible-collections/ansible.mcp (10 stars, 147 commits, GPL-3.0, created September 2025) is the official Ansible MCP Collection — not an MCP server itself, but Ansible plugins that let playbooks interact with MCP servers. Playbooks can dynamically discover available MCP servers, retrieve tool lists, and execute tools from within automation workflows. This is the complement to mcp_builder: where mcp_builder installs MCP servers into Execution Environments, ansible.mcp lets playbooks consume them.
bsahane/mcp-ansible (30 stars, Python) is the most feature-rich community server. 35+ tools across playbook execution, inventory management (parse, graph, diff, find-host), project management, vault operations (encrypt, decrypt, view, rekey), troubleshooting (remote commands, log fetching, service management), diagnostics (health monitoring, performance baselines, state comparison), and advanced analysis (network matrix, security audit, auto-heal). The “Swiss Army knife” for Ansible operations.
sibilleb/AAP-Enterprise-MCP-Server (31 stars, Python) targets enterprise AAP + EDA environments. 50+ tools covering AAP inventory/host/job/project management, Ansible Galaxy search and recommendations, Event-Driven Ansible (activation management, rulebook querying, event monitoring), Ansible Lint quality tools, and Red Hat documentation access with secure domain validation. This is the enterprise-grade community option.
tarnover/mcp-sysoperator (26 stars, TypeScript, MIT) combines Ansible with Terraform and LocalStack in a single server. Ansible tools for playbook execution, inventory management, and vault operations. Terraform tools for init/plan/apply/destroy. AWS tools for EC2/S3/VPC/CloudFormation. LocalStack integration for testing without real AWS credentials.
Correction (2026-08-15 audit): the repo was archived by its owner on May 27, 2026 and is now read-only — this review previously described it as “still in active development.” No successor project is referenced in the archival notice.
Red Hat also provides ansible.mcp_builder (3 stars, 77 commits, v1.0.3 January 2026) — an Ansible Collection that installs MCP servers into Execution Environments (EEs) from npm, PyPI, or compiled Go binaries, with a unified registry system and auto-generated manifest files.
Pulumi
| Server | Stars | Language | Tools | Transport |
|---|---|---|---|---|
npm: @pulumi/mcp-server (source no longer public) | — | TypeScript | 15+ | stdio, HTTP |
Remote hosted (mcp.ai.pulumi.com/mcp) | — | — | 12+ | HTTP (OAuth) |
Pulumi pushes infrastructure MCP further than any other vendor with autonomous infrastructure provisioning via Neo delegation.
Correction (2026-08-15 audit): a prior version of this review cited the pulumi/mcp-server GitHub repo at ~188 stars with an Apache-2.0 license. That repo no longer resolves — github.com/pulumi/mcp-server now 404s, and searching the pulumi org on GitHub for “mcp” returns zero results. The npm package @pulumi/mcp-server still exists but hasn’t shipped a release since v0.2.0 on 2025-09-26. Pulumi’s own current MCP server docs now describe the product as a remote-only hosted service at mcp.ai.pulumi.com/mcp with no mention of a local/open-source install — Pulumi appears to have consolidated the local mode into the hosted offering and taken the source repo private or removed it. The star count and license claim above could not be re-verified and are cut.
Per Pulumi’s current documentation, the server now operates as a single hosted service:
Registry & deployment tools — get-type, get-resource, get-function, list-resources, list-functions, and the deploy-to-aws quickstart tool.
Pulumi Cloud tools — get-stacks (list stacks), resource-search (query across cloud resources in your organization), get-policy-violations (compliance checking), get-users (org members).
The standout feature is still Neo delegation, now folded into the hosted server: neo-task-launcher launches Pulumi Neo — an autonomous AI agent that handles multi-step infrastructure tasks end to end — with neo-get-tasks to monitor progress and neo-continue-task/neo-reset-conversation to manage the interaction. This is infrastructure delegation, not just tool calling. See the Pulumi remote MCP server announcement for the rationale behind going remote-only (zero local setup, centralized OAuth auth, automatic updates).
OAuth authentication required. One-click install for Cursor, Claude Code CLI integration (claude mcp add --transport http pulumi https://mcp.ai.pulumi.com/mcp), Claude Desktop, Windsurf, and Kiro.
OpenTofu
| Server | Stars | Language | Tools | Transport |
|---|---|---|---|---|
| opentofu/opentofu-mcp-server | ~107 | TypeScript | 5 | stdio, HTTP |
OpenTofu’s official MCP server (107 stars) mirrors Terraform’s documentation-focused philosophy — registry access only, no execution.
5 tools:
| Tool | What it does |
|---|---|
search-opentofu-registry | Search for providers, modules, resources, data sources |
get-provider-details | Comprehensive provider documentation |
get-module-details | Detailed module specifications |
get-resource-docs | Resource-specific documentation |
get-datasource-docs | Data source documentation |
v1.0.0 (June 6, 2026) — the project’s first stable release — MPL-2.0 licensed. Available as a hosted service at mcp.opentofu.org or locally via npx @opentofu/opentofu-mcp-server. Cloudflare Worker deployment option for self-hosted remote.
Functionally similar to Terraform’s registry tools but for the OpenTofu ecosystem. If you’ve migrated from Terraform to OpenTofu, this is the direct replacement — though it lacks workspace management, policy tools, and the broader toolset coverage of HashiCorp’s server.
Crossplane / Upbound
| Server | Stars | Language | Tools | Status |
|---|---|---|---|---|
| upbound/marketplace-mcp-server | ~6 | Go | 9 | Active |
| vfarcic/crossplane-mcp | ~1 | Go | 1 | MVP |
| cychiang/crossplane-mcp-server | ~1 | Python | 4+ | Archived |
Crossplane’s MCP presence improved significantly with Upbound’s marketplace server.
upbound/marketplace-mcp-server (6 stars, 43 commits, Go) is the first official Crossplane-ecosystem MCP server from Upbound, the company behind UXP 2.0 (AI-native Crossplane distro). 9 tools for searching, discovering, and managing packages in the Upbound Marketplace: search_packages, get_package_metadata, get_package_assets, get_repositories, reload_auth, get_package_version_resources, get_package_version_composition_resources, get_package_version_groupkind_resources, get_package_version_examples. Supports HTTP and stdio transport. Requires UP CLI authentication.
vfarcic/crossplane-mcp provides a single ListClaimsBasic tool to list Claim names and namespaces for a configured API group. MVP only — plans for service creation, observation, and deletion exist but aren’t implemented yet.
cychiang/crossplane-mcp-server was a Python implementation for querying Crossplane resources (XRDs, Compositions, ManagedResources) but the repository is archived as of September 2025.
The Kubernetes MCP server can also interact with Crossplane resources via standard kubectl operations.
Terraform Cloud Alternatives
| Server | Stars | Language | Tools | Transport |
|---|---|---|---|---|
| env0/mcp-server | ~4 | TypeScript | 10+ | stdio |
| spacelift-io/spacelift-intent | ~137 | — | 15+ | Hosted |
env0 launched an official MCP server (4 stars, 46 commits, TypeScript) — connecting AI agents to env0’s platform for environment deployment, cancellation, and log retrieval, plus Cloud Compass integration for retrieving cloud resource configurations with advanced filtering. Can generate Terraform/OpenTofu code from existing cloud resources. Compatible with Cursor, Claude Code, VS Code, Windsurf, Cline, Zed, JetBrains, and others. env0’s platform also includes Cloud Analyst (AI-powered infrastructure insights via natural language) and Code Optimizer (beta, IaC security/config scanning with actionable Git fixes).
Spacelift now has MCP presence via two offerings. First, spacelift-io/spacelift-intent (137 stars, Apache-2.0) — an MCP server for natural language infrastructure provisioning that calls provider APIs directly, bypassing IaC files entirely. The current README documents 15+ tools across provider discovery/schema, resource lifecycle (create/update/delete/refresh/import), data sources, state management, and dependency management (down from the 18 previously reported here — cut to match the live tool list). Second, Spacelift accounts expose a hosted MCP endpoint (/mcp at your account URL) with 5 tools: discover, query, mutate, provider, and intent — giving Claude agents access to Spacelift’s GraphQL API directly. The previously flagged gap is now filled.
Scalr still has no MCP presence.
Cost Estimation
| Server | Stars | Language | Tools | Transport |
|---|---|---|---|---|
| phildougherty/infracost_mcp | ~2 | TypeScript | 16 | stdio |
phildougherty/infracost_mcp (2 stars, 10 commits, MIT) is the first Infracost MCP server — filling a gap we flagged in the previous version of this review. 16 tools: cost breakdown generation (JSON, HTML, table, diff formats), configuration comparison (diff), Infracost Cloud upload for centralized tracking, PR commenting (GitHub, GitLab, Azure Repos, Bitbucket), tagging policy management (create/list/get/update/delete with ANY/LIST/REGEX validation), cost guardrail management (thresholds that block PRs), and usage YAML templates (small/medium/large). Early-stage but functional.
Drift Detection
| Server | Stars | Language | Tools | Transport |
|---|---|---|---|---|
| terramate-io/terramate-mcp-server | ~5 | — | 13 | stdio |
terramate-io/terramate-mcp-server (5 stars) is the first MCP server with dedicated IaC drift detection — filling a gap we flagged in the April review. 13 tools covering Terramate Cloud operations: stack management, drift detection (view drift runs and retrieve Terraform plan outputs for drifted resources), pull request integration, deployment tracking, and stack resource listing. This fills a meaningful monitoring gap — where tfmcp’s analyze_state approach requires direct CLI access, Terramate’s server connects to cloud-managed drift monitoring. Still very early-stage (5 stars), but the category needed exactly this.
Multi-Tool / Cross-Platform
| Server | Stars | Language | Coverage |
|---|---|---|---|
| tarnover/mcp-sysoperator | ~26 | TypeScript | Ansible + Terraform + AWS (archived) |
MCP-sysoperator (covered under Ansible above) is the only server attempting cross-platform IaC coverage. Combining Ansible playbook execution with Terraform plan/apply and LocalStack testing in a single MCP server is genuinely useful for teams using both tools — though 26 stars suggested limited adoption even before the repo was archived on May 27, 2026.
What’s Missing
- No Chef, Puppet/OpenVox, or SaltStack MCP servers — Chef Infra Server hits end-of-life November 2026, Puppet’s community forked to OpenVox after Perforce restricted binary access, and SaltStack remains quiet. Zero MCP presence across all three, reflecting the industry’s accelerating shift toward IaC and immutable infrastructure
- No cross-IaC registry — no single server queries Terraform Registry, Pulumi Registry, and OpenTofu Registry simultaneously
- No Scalr MCP server — Scalr remains the only major Terraform Cloud alternative with no MCP presence
No Spacelift MCP server— gap filled: spacelift-io/spacelift-intent (137 stars) for natural language provisioning + hosted MCP endpoint at Spacelift account URLNo drift detection MCP— gap filled: terramate-io/terramate-mcp-server (5 stars, 13 tools) adds cloud-managed drift monitoring via Terramate CloudNo CDK MCP server— gap filled (covered via AWS MCP collection‘s dedicatedcdk-mcp-serversub-server)No cost estimation MCP— gap filled: phildougherty/infracost_mcp now provides 16 tools for Terraform cost estimation, though still early-stage (2 stars)No Crossplane official server— gap filled: upbound/marketplace-mcp-server provides 9 tools for Crossplane package discoveryNo env0 MCP server— gap filled: env0 shipped an official MCP server with Cloud Compass integration
The Bottom Line
Rating: 4.5 / 5 — The category has matured significantly. HashiCorp’s Terraform MCP server (~1,500 stars, v1.2.0) is now firmly in execution territory — plan/apply, Stacks, policy management, and OpenTelemetry instrumentation. Red Hat open-sourced ansible/aap-mcp-server (32 stars) with full multi-service AAP coverage and Prometheus metrics, though it remains a Technology Preview rather than GA. Two additional previously-flagged gaps are now filled: Spacelift shipped spacelift-io/spacelift-intent (137 stars) for natural language provisioning plus a hosted GraphQL MCP endpoint, and Terramate launched a dedicated drift detection MCP server. The category loses half a point for the configuration management gap (Chef EOL November 2026, no Puppet/OpenVox or SaltStack MCP servers) and the lack of a cross-IaC registry — but these reflect declining market relevance rather than missed opportunities.
Best for Terraform teams: hashicorp/terraform-mcp-server for registry intelligence + plan/apply (v1.2.0), paired with severity1/terraform-cloud-mcp for full TFC API coverage or nwiizo/tfmcp (v0.2.2) for direct CLI access.
Best for Ansible teams: ansible/aap-mcp-server (official, open-source, Technology Preview) for AAP 2.6.4+ deployments willing to run pre-GA software. For enterprise environments, sibilleb/AAP-Enterprise-MCP-Server (50+ tools) adds EDA and Galaxy AI recommendations. For standalone Ansible, bsahane/mcp-ansible (35+ tools) covers the most operational ground.
Best for Pulumi teams: Pulumi’s MCP server — now remote-hosted only — with Neo delegation is unique in the IaC space.
Best for Spacelift teams: spacelift-io/spacelift-intent for natural language infrastructure provisioning, or the hosted GraphQL endpoint for account-level operations.
Best for cost management: phildougherty/infracost_mcp for Terraform cost estimation — early-stage but fills a critical gap.
Best for drift monitoring: terramate-io/terramate-mcp-server for Terramate Cloud environments — the first dedicated drift detection MCP server in the IaC space.
This review covers publicly available information as of August 2026. ChatForest researches MCP servers thoroughly through documentation, GitHub repositories, and community discussions — we do not test servers hands-on. Star counts are approximate and change over time. Always check the linked repositories for the latest status.
This review was last edited on 2026-08-15 using Claude Sonnet 5 (Anthropic) for a claim-level citation audit: verified star counts, license, and version numbers for every listed server against live GitHub/npm data; corrected the ansible/aap-mcp-server GA claim (it is Technology Preview per Red Hat’s own June 2026 documentation); flagged tarnover/mcp-sysoperator as archived; and rewrote the Pulumi section after the pulumi/mcp-server GitHub repo was found to no longer resolve (Pulumi has moved to a remote-only hosted MCP server).