Atlas drives Google Cloud through the official gcloud MCP server, which wraps the gcloud CLI and inherits your active account's IAM permissions rather than taking a separate API key. Wrap the CLI in MCP with atlas mcp add gcloud -- npx -y @google-cloud/gcloud-mcp, which needs Node 20 or newer and the gcloud CLI already installed, then authenticate with gcloud auth login.
How Atlas connects to Google Cloud
Atlas drives Google Cloud through the official gcloud MCP server, which wraps the gcloud CLI rather than talking to a separate API. In 2026 that design has one large consequence: the server inherits your active account's IAM permissions instead of taking a dedicated API key, so Atlas is bounded by exactly what you are.
Google Cloud is a cloud platform, and the gcloud MCP server is a thin wrapper around the gcloud CLI you already have installed. That is deliberate. Rather than issuing Atlas its own credential with its own lifecycle and its own audit trail, the server borrows the identity of the gcloud account that is currently active on your machine. If your account cannot read a Cloud Run revision, neither can Atlas. If your account can deploy, Atlas can propose a deploy, and you still approve it. Permissions review stays a Google Cloud IAM problem, which is where your organization already knows how to solve it.
Installing the gcloud MCP server for Atlas
Wrap the gcloud CLI in MCP with atlas mcp add gcloud -- npx -y @google-cloud/gcloud-mcp. The server needs Node 20 or newer and the gcloud CLI already installed on the machine. If you would rather not hand-edit atlas.json, npx @google-cloud/gcloud-mcp init scaffolds the client wiring for you.
Two prerequisites decide whether the install works on the first try: Node 20 or newer, and a gcloud CLI that is already present. The MCP server does not bundle gcloud, it shells out to it, so a machine without the CLI will produce confusing tool errors rather than a clean failure. Check both before you add the server. Once the prerequisites are in place, atlas mcp add gcloud -- npx -y @google-cloud/gcloud-mcp registers it, or you can run npx @google-cloud/gcloud-mcp init and let it scaffold the client wiring into atlas.json rather than editing the file by hand.
Authenticating Atlas to Google Cloud
Authentication for the gcloud MCP server is the active gcloud account, not an API key. Run gcloud auth login and gcloud auth application-default login, because there is no dedicated API key env var for Atlas to read. In 2026 this is the whole of the credential setup for Google Cloud.
Developers coming from other cloud integrations look for the environment variable and do not find one. There is not supposed to be one. The gcloud MCP server wraps the gcloud CLI, and the gcloud CLI authenticates with your account. Run gcloud auth login for the CLI session and gcloud auth application-default login for the application default credentials that client libraries use. Both matter. When the two are set, Atlas inherits the resulting IAM permissions automatically, and revoking Atlas's Google Cloud access is as simple as revoking the gcloud account it is borrowing.
Debugging a failing Cloud Run revision with Atlas
The signature Atlas and Google Cloud workflow in 2026 is a Cloud Run triage. Ask Atlas to list Cloud Run revisions, read the failing one's logs, and find the handler that returns 500. Because the gcloud MCP server wraps the gcloud CLI, Atlas gathers all three in a single pass.
A Cloud Run 500 has a shape: some revision is bad, its logs contain the trace, and the trace points at a handler in the repository you are sitting in. Atlas walks that chain without you switching windows. It lists the revisions to find which one is serving the errors, reads that revision's logs to get the trace, and then opens the handler in your working tree. The last step is what a log console cannot do. Atlas has the source, so it can tell you not just that the handler returns 500 but why the code path gets there.
Letting Atlas patch the Dockerfile and cloudbuild.yaml
Atlas can patch the Dockerfile and cloudbuild.yaml, then wait for you to approve gcloud run deploy when it prompts. In 2026 this keeps Google Cloud deploys on the same review model as any other Atlas change: read the diff first, run the deploy second, and never in the other order.
Build configuration is where Cloud Run problems often actually live. A missing system package in the Dockerfile, a build step in cloudbuild.yaml that stopped producing the artifact the service expects. Both are files in your repository, so Atlas edits them the way it edits any file: as a diff you read. The deploy itself stays a separate, explicit act. When Atlas is ready to run gcloud run deploy, it prompts, and you approve. Because the gcloud MCP server inherits your IAM permissions, the deploy runs as you, with your access, and shows up in the audit log as such.
Adding managed Google Cloud endpoints alongside gcloud
Google also runs managed per-service MCP endpoints, so you can add https://bigquery.googleapis.com/mcp alongside the gcloud MCP server for data work. In 2026 Atlas can hold both connections at once: the gcloud wrapper for platform operations, and the managed BigQuery endpoint for querying.
The gcloud MCP server is general purpose because gcloud is general purpose, but data work has a sharper tool. Google's managed per-service endpoints expose individual products directly, and https://bigquery.googleapis.com/mcp is the one most teams reach for. Adding it does not replace the gcloud MCP server. The two coexist, and Atlas picks the right one for the request: gcloud when you are listing Cloud Run revisions or deploying, BigQuery when you are exploring a dataset. Set them both up once and stop thinking about which is which.
Setup
- 01Confirm the prerequisites: Node 20 or newer, and the gcloud CLI already installed on the machine, since the MCP server shells out to it.
- 02Wrap the gcloud CLI in MCP with atlas mcp add gcloud -- npx -y @google-cloud/gcloud-mcp
- 03Authenticate with the active gcloud account: run gcloud auth login and gcloud auth application-default login, because there is no dedicated API key env var.
- 04If you would rather not hand-edit atlas.json, run npx @google-cloud/gcloud-mcp init to scaffold the client wiring.
- 05Ask Atlas to list Cloud Run revisions, read the failing one's logs, and find the handler that returns 500.
- 06Let Atlas patch the Dockerfile and cloudbuild.yaml, then approve gcloud run deploy when it prompts.
- 07For data work, add the managed endpoint https://bigquery.googleapis.com/mcp alongside the gcloud MCP server.
Frequently asked questions
- how to connect atlas to google cloud
- Run atlas mcp add gcloud -- npx -y @google-cloud/gcloud-mcp. You need Node 20 or newer and the gcloud CLI already installed, then authenticate with gcloud auth login.
- what api key does the gcloud mcp server need
- None. Authentication is the active gcloud account, so run gcloud auth login and gcloud auth application-default login. There is no dedicated API key env var.
- does the gcloud mcp server require node
- Yes. The gcloud MCP server needs Node 20 or newer and the gcloud CLI already installed, because it wraps the CLI rather than replacing it.
- what iam permissions does atlas have on google cloud
- Exactly your own. The gcloud MCP server inherits your active account's IAM permissions rather than taking a separate API key, so Atlas cannot exceed what you can do.
- can atlas debug a failing cloud run revision
- Yes. Ask Atlas to list Cloud Run revisions, read the failing one's logs, and find the handler that returns 500. It then opens that handler in your working tree.
- how do i set up the gcloud mcp server without editing atlas.json
- Run npx @google-cloud/gcloud-mcp init. It scaffolds the client wiring for you instead of requiring a hand-edit of atlas.json.
- can atlas query bigquery
- Yes. Google runs managed per-service endpoints, so add https://bigquery.googleapis.com/mcp alongside the gcloud MCP server for data work.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas with Baseten in 2026: 262,000 Tokens In, 262,000 Tokens Out
Atlas with Baseten in 2026: Kimi K2.7 Code at 262,000 tokens in and 262,000 out, GPT OSS 120B at $0.10/$0.50 per Mtok, and the tradeoffs of dedicated hosting.
Research a Third-Party API Before Integrating It with Atlas in 2026
How to research a third-party API with Atlas in 2026: websearch finds the current docs, webfetch pulls the page as markdown or text, and grep checks repo conventions.
Atlas for Astro: Islands, Content Collections, and Zero JS by Default in 2026
Atlas is a terminal-native AI coding agent for Astro in 2026. It reads astro.config.mjs, src/pages, and content collection schemas, drops needless client:load directives, and runs astro check.
Atlas with OpenAI o4-mini (2026): Cheap Reasoning for Parallel Subagents
OpenAI o4-mini drives Atlas at $1.10 per Mtok input and $4.40 per Mtok output on a 200K context. A strict upgrade over o3-mini at identical price, with real tradeoffs.
Atlas with IBM Granite 4 Small-H (Ollama): a 1M-Token Local Window in 2026
IBM Granite 4 Small-H (Ollama) is a hybrid Mamba model with 1M tokens (1,048,576) of context from a 19GB download, Free (self-hosted). Atlas setup and memory notes.
Atlas with GPT-5.4 Pro: One Shot Hard Problems in 2026
GPT-5.4 Pro is the max reasoning tier at $30 per Mtok input, $180 per Mtok output on a 1,050,000 token window. Why Atlas users switch to it instead of pinning it.
Atlas for Haskell in 2026
Atlas is a terminal-native AI coding agent for Haskell in 2026. Run it in a project with a .cabal file or a stack.yaml, let cabal build drive it, and review the diff.
Atlas with Magistral Medium: Multi-Hop Root-Cause Debugging in 2026
Magistral Medium reasons across a 128,000 token window at $2.00 / 1M input tokens and $5.00 / 1M output tokens. Atlas setup, the 16,384 token output cap, tradeoffs.