OpsCanvas MCP Operator Edition

The safe, grounded way to improve your cloud with AI.

First released September 1, 2026 · Free for individual use · No account required to start

AI agents are only as accurate as their context. OpsCanvas gives them real context on your cloud.

OpsCanvas builds your Cloud Intelligence Graph™ across your AWS and Azure estate - topology, dependencies, and cost, all mapped without tagging. Only data you've confirmed makes it into the graph. Blast radius isn't stored as a map inside it; it's a question the graph can answer for any change, on demand.

Connect the AI you already use and that graph is what grounds it. It reads from your graph to make better-informed decisions, and only writes back finalized data you've confirmed. What it actually does out in your cloud, and how much autonomy you give it, is between you and your model - OpsCanvas isn't in that decision.

Setup registers OpsCanvas with every supported harness it finds on the machine:

• [x] Claude Code (CLI and desktop)
• [x] Codex
• [x] Cursor
• [x] Antigravity
• [x] OpenCode
• [ ] Grok Build
• [x] Copilot CLI

OpsCanvas isn't another agent you switch to. There's no new chat window, no destination to learn. It's the grounding layer underneath the AI you're already running.

What's in the first release

This package is the Operator edition: the MCP server you install on your own machine. Pro and Enterprise build on the same graph and are not part of this package.

What's coming next

The full Organize → Prioritize → Fix → Verify improvement loop:

• Organize. Findings from your FinOps tools, your cloud providers' native tools, and OpsCanvas's own built-in checks, landing in one queue.
• Prioritize. Every finding enriched with what the evidence actually supports - blast radius, effort, cost impact - then ranked by benefit weighed against effort and risk, with the reasoning shown. Not a sorted list: a defensible "start here."
• Fix. Where OpsCanvas has the context and confidence, your AI gets a remediation plan and walks the execution with you - guardrailed, step by step. Where it doesn't, you get a straight answer about why, instead of a fabricated plan.
• Verify. Confirmation that the improvement actually landed. Not "recommendation sent" - savings confirmed.

Also on the roadmap, in no committed order: shared queues and task assignment for teams; two-way integrations with the FinOps and ticketing tools a team already runs; the same loop extended to resilience and risk findings, including backup platforms; and security findings, with cross-system dependency awareness so one remediation can close findings in more than one system.

Why it's safe

• Local data plane, hosted control plane. Your credentials never leave your machine. Only graph metadata and coordination state sync.
• Read-only by default, for building the graph. Mapping your estate doesn't touch or change anything. What your AI does with what it reads - and how much autonomy you give it - is a decision between you and your model; OpsCanvas isn't in the loop for that.
• You confirm what goes in. Only data you've reviewed and confirmed becomes part of the graph, and only finalized data you've blessed gets written back to it.
• Free to explore, gated to connect. Install, connect your cloud, and view and confirm your topology without ever typing an email address. Registering is what makes that graph available for your AI to read - you'll see that gate clearly before you hit it, and your AI will tell you too.

Editions

| | Operator | Pro | Enterprise | |---|---|---|---| | Who it's for | One person | A team | A function | | The Cloud Intelligence Graph™ | ✓ | ✓ | ✓ | | Remediation loop (plan, approve, apply, verify, revert) | Roadmap - all tiers | Roadmap - all tiers | Roadmap - all tiers | | Assessment depth | Directional | Directional | Deep, continuous | | Shared queue, task assignment | – | ✓ | ✓ | | Continuous monitoring, live integrations | – | – | ✓ | | Deployment | Hosted | Hosted | Self-hosted option |

Operator is free. See opscanvas.ai/pricing for what Pro and Enterprise cost, which is kept there rather than here: a published version of this page cannot be withdrawn, and a price should not be frozen into one.

An individual workspace rolls up into a team's when you join one - nothing is exported or migrated. Moving from Pro to Enterprise is a repoint, not a rebuild.

Quick start

Two commands. Installing places the files; setup is what registers OpsCanvas and starts it.

npm install -g @opscanvas/opscanvas
opscanvas setup

setup registers OpsCanvas with whichever supported harnesses are on the machine, starts its background process, and asks where your repositories live and which cloud identities to use. It runs at your terminal, so it can ask; an install cannot.

Upgrading is the same two commands, except on Windows (below). Re-running setup is what replaces a daemon still serving from the version you just replaced. Skip it and your next session will tell you, but it will have to fail once first.

On Windows, run opscanvas uninstall before upgrading. Windows will not let anything replace a running executable, and nothing OpsCanvas ships runs before your package manager does. Upgrade with the background process running and npm reports success while leaving the old version serving, and the upgrade after that fails outright until the process is stopped.

The first opscanvas command fetches and checksum-verifies the one binary your platform needs. OPERATOR_BINARY_BASE_URL, HTTPS_PROXY and NO_PROXY all apply to that fetch the way they would to any other download this machine makes.

Node 22.13 or newer. Binaries ship for macOS on Apple silicon and Intel, Linux on arm64 and x64, and Windows on x64; there is no Windows arm64 build yet, and OpsCanvas reports that rather than guessing.

If the background process is not running

opscanvas service status     # what your platform's own service manager says
opscanvas service restart    # get a daemon running now

service restart is what a failed tool call tells you to run, and it works whether or not anything is registered. Whether it comes back on its own at the next login is a separate question, answered by re-running setup, not by service; on a machine with no user-level service manager it cannot come back on its own at all, and restart after each reboot is the answer rather than a workaround.

Removing OpsCanvas

Two commands, in this order:

opscanvas uninstall
npm uninstall -g @opscanvas/opscanvas

The order matters, and the second command cannot make up for skipping the first. npm removes files; it does not stop a running process, and it runs no uninstall script a package could hook. Remove the package on its own and the daemon keeps serving from a binary that is no longer on disk, the service registration keeps trying to start it at every login, and the hooks and skill that setup installed keep running in every session of every harness.

opscanvas uninstall is what undoes all of that: it unregisters from each harness, removes the hooks, the approval and the installed skill, stops the daemon, and deletes the service registration. It leaves your data where it is and tells you where that is; nothing else will delete it.

To confirm the package itself is gone, npm -g list lists what is still installed globally, and @opscanvas/opscanvas (or @therealopscanvas/opscanvas on the private channel) should no longer be in it. That is a check rather than a third step: skipping it removes nothing and leaves nothing running.

Support

The browser app has a feedback form. Submitting it opens a support ticket with OpsCanvas support and sends your installation id, version and platform along with the text you write.

License

GPL-3.0-only.