Atlas works with Nx through Nx's official MCP server, which hands Atlas the real project graph. Add it with atlas mcp add nx -- npx nx mcp, or run npx nx configure-ai-agents to bootstrap the wiring if you would rather not hand-edit atlas.json. With the graph in hand, Atlas knows which packages a change actually affects instead of inferring it from import statements, and can run nx affected on exactly those projects rather than testing the whole monorepo.
What the Atlas and Nx integration does
Atlas integrates with Nx through Nx's official MCP server, which hands Atlas the real project graph. In 2026 that means Atlas knows which packages a change actually affects instead of inferring it from import statements, and can run nx affected on exactly those projects.
The project graph is the whole point of the Nx integration. In a large monorepo, working out the blast radius of a change from import statements alone is unreliable: implicit dependencies, generated code, and config-level relationships do not show up in an import scan, and an agent that guesses will either test too much or miss the package it broke. Nx already computes the true graph, and Nx's official MCP server hands that graph to Atlas directly. So when Atlas says a change touches four projects, it is reporting what Nx computed, not what it inferred. Everything else in this integration follows from that one fact, including the ability to run nx affected on exactly the projects a change actually touched.
How to add the Nx MCP server to Atlas
Adding Nx to Atlas is a single command: atlas mcp add nx -- npx nx mcp. On Nx below 21.4, use npx nx-mcp@latest instead. Either way the Nx MCP server starts alongside Atlas and exposes the workspace project graph to the agent.
Run atlas mcp add nx -- npx nx mcp from your Nx workspace and Atlas registers Nx's official MCP server. The version check matters before you run it: on Nx below 21.4 the mcp subcommand is not there, so use npx nx-mcp@latest instead, which pulls the standalone package. Both paths land in the same place, an Nx MCP server that Atlas can query for the project graph. Check the Nx version in your workspace first rather than debugging a failed command afterwards. Once the server is registered, Atlas has Nx tools available in the session and can answer questions about the workspace structure using the graph rather than reading through package manifests one at a time.
Bootstrapping the wiring with npx nx configure-ai-agents
Running npx nx configure-ai-agents bootstraps the wiring for Atlas if you would rather not hand-edit atlas.json. The Nx command sets up the MCP server configuration for you, which is the fastest path in a workspace where several developers all need the same Nx MCP setup in 2026.
There are two ways to get the Nx MCP server into Atlas. The explicit one is atlas mcp add nx -- npx nx mcp, which is a single command and easy to read in a README. The other is npx nx configure-ai-agents, which bootstraps the wiring if you would rather not hand-edit atlas.json. The second option is usually the better one for a team, because it is a command any developer can run on a fresh clone to arrive at the same configuration, with no copy-pasted JSON drifting between machines. Pick whichever fits how your workspace onboards people. The result Atlas sees is identical: Nx's official MCP server, registered and ready to serve the project graph.
Minimal mode and the Nx tools it hides
The Nx MCP server runs in --minimal mode by default, which hides 2 tool groups: the workspace-analysis and generator tools. Pass --no-minimal to bring them back when you want Atlas to use Nx generators or dig into workspace analysis rather than only reading the project graph.
Minimal mode is the default because a smaller tool surface keeps an agent focused, and for most work the project graph is all Atlas needs. When you want more, pass --no-minimal and the workspace-analysis and generator tools come back. Generators are the obvious reason to reach for it: if you want Atlas to scaffold a new library or app the way your workspace expects, it needs the Nx generator tools, and in --minimal mode they are hidden. Workspace analysis is the other. Start in the default --minimal mode, and switch to --no-minimal when a task genuinely calls for generation or deeper analysis. Knowing which mode you are in explains most cases of a missing Nx tool in an Atlas session.
The daily workflow: nx affected -t test on only what changed
The core Atlas plus Nx loop is 1 request: ask Atlas to run nx affected -t test and fix only the projects the change actually touched. Because the Nx MCP server hands Atlas the real project graph, the affected set is computed by Nx, not guessed from imports.
In a monorepo the expensive habit is running the entire test suite because nobody is sure what a change broke. With Nx's official MCP server registered, ask Atlas to run nx affected -t test and it scopes the run to the projects the change actually touched. Atlas then works through the failures in exactly those projects. The feedback loop stays short, which matters more than it sounds: a fast affected run is one Atlas can repeat after every fix, so the agent iterates against real test results instead of reasoning about what might break. That loop is only trustworthy because the affected set comes from the real project graph rather than an import scan, which is what the Nx MCP server exists to provide.
Nx Cloud tools in Atlas
Nx Cloud tools appear automatically in Atlas once the workspace is connected, so there is no extra token to add. Developers already running Nx Cloud on their monorepo get those tools through the same atlas mcp add nx -- npx nx mcp server they registered in 2026.
Nx Cloud requires no separate setup on the Atlas side. Once the workspace is connected to Nx Cloud, the Nx Cloud tools appear automatically in the Nx MCP server that Atlas is already talking to, and there is no extra token to add to atlas.json. That is worth calling out because the usual assumption with any cloud service is a second credential and a second registration step, and here there is neither. If your workspace is connected, the tools are there. If it is not, the rest of the integration still works exactly as described: the project graph, nx affected -t test, and the choice between --minimal and --no-minimal are all local to the workspace.
Setup
- 01Add Nx's official MCP server with atlas mcp add nx -- npx nx mcp from the root of your Nx workspace.
- 02On Nx below 21.4, use npx nx-mcp@latest instead, since the nx mcp subcommand is not available on those versions.
- 03Alternatively run npx nx configure-ai-agents, which bootstraps the wiring if you would rather not hand-edit atlas.json.
- 04Leave the server in --minimal mode for project graph work, since it is the default and hides the workspace-analysis and generator tools.
- 05Pass --no-minimal when you want Atlas to use the Nx generator or workspace-analysis tools.
- 06Confirm Atlas has the real project graph by asking it which packages a given change affects.
- 07Ask Atlas to run nx affected -t test and fix only the projects the change actually touched.
- 08Connect the workspace to Nx Cloud and the Nx Cloud tools appear automatically, with no extra token to add.
Frequently asked questions
- how do I connect Atlas to an Nx monorepo
- Run atlas mcp add nx -- npx nx mcp from the workspace root to register Nx's official MCP server. On Nx below 21.4, use npx nx-mcp@latest instead.
- what does the Nx MCP server give Atlas
- The real project graph. Atlas knows which packages a change actually affects instead of inferring it from import statements, and can run nx affected on exactly those projects.
- why are Nx generator tools missing in Atlas
- The Nx MCP server runs in --minimal mode by default, which hides the workspace-analysis and generator tools. Pass --no-minimal to bring them back.
- how do I set up Nx MCP without editing atlas.json by hand
- Run npx nx configure-ai-agents. It bootstraps the wiring for you, which is easier than hand-editing atlas.json and gives every developer on the workspace the same setup.
- can Atlas run only the affected Nx tests
- Yes. Ask Atlas to run nx affected -t test and it fixes only the projects the change actually touched, because the affected set comes from the real project graph.
- do I need an Nx Cloud token to use Nx Cloud tools in Atlas
- No. Nx Cloud tools appear automatically once the workspace is connected, so there is no extra token to add.
- what Nx version do I need for the nx mcp command
- The nx mcp subcommand needs Nx 21.4 or later. On Nx below 21.4, use npx nx-mcp@latest instead.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas with DeepSeek V3.1 (open weights): Togglable Thinking in 2026
Run Atlas on DeepSeek V3.1 (open weights) in 2026. One MIT-licensed checkpoint with thinking and non-thinking modes, $0.25 per Mtok in and $0.95 per Mtok out.
Atlas with DeepSeek V3.2 (open weights): Sparse Attention at $0.38 Output, 2026
Run Atlas on DeepSeek V3.2 (open weights) in 2026. DeepSeek Sparse Attention gives 160K tokens (DeepInfra) at $0.26 per Mtok in and $0.38 per Mtok out.
Atlas with GPT-5.6 Terra: The Mid Tier GPT-5.6 Pick for 2026
GPT-5.6 Terra drives Atlas on a 1,050,000 token context at $2.50 per Mtok input, $15 per Mtok output. Setup, cost math, and when Sol or Luna is the better pick.
Atlas with Claude Opus 4.8: The 2026 Model Guide
Claude Opus 4.8 in Atlas: a 1M token context window at $5 / $25 per Mtok with 128K output. When to pin Anthropic's top coding model in 2026, and when not to.
Atlas with GLM-4.5-Flash: The Free Reasoning Slot in 2026
GLM-4.5-Flash is listed at $0.00 per Mtok input and output on Z.ai, with the full 128K tokens (131,072) context. Free background traffic for Atlas, rate limited.
Atlas with Grok Build 0.1: xAI's First Coding-Agent Model (2026)
Grok Build 0.1 runs in Atlas at $1 / $2 per Mtok with a 256K context and 256,000 max output tokens. A coding-specialized model, but it is a 0.1 release.
Atlas vs Cursor: terminal AI coding agents compared (2026)
A grounded 2026 comparison of Atlas and Cursor across workflow, change review, extensibility, and pricing for developers choosing an AI coding agent.
Atlas with GPT-5 Mini: Full 400K Context at One Fifth the Price in 2026
GPT-5 Mini in Atlas: $0.25 per Mtok input and $2 per Mtok output, 5x cheaper than GPT-5 on both sides, with no reduction to the 400K context window.