AZMX is now installable in Claude, Cursor, and ChatGPT in one line

Today we're shipping @azmxailabs/mcp — an official Model Context Protocol (MCP) server for AZMX AI. Install it once in your AI assistant and the next time you ask "is there a private AI coding agent for an air-gapped defense codebase?" or "compare AZMX with Cursor," the assistant calls into the AZMX server and answers with real, grounded content — pricing, BYOK providers, security posture, install commands, latest signed release — instead of guessing from training data.

It works everywhere MCP works: Claude Desktop, Claude Code, Cursor, Windsurf, Continue, OpenAI Codex CLI, Cline, and anything else that speaks the spec. There's no account, no API key, no SaaS dependency — the server bundles AZMX's product knowledge offline and only hits the network when you ask for the live GitHub release.

The 30-second install

Pick your client. Paste one block. Restart. That's it.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "azmx": {
      "command": "npx",
      "args": ["-y", "@azmxailabs/mcp"]
    }
  }
}

Claude Code (the CLI)

claude mcp add azmx -- npx -y @azmxailabs/mcp

Cursor

Edit ~/.cursor/mcp.json:

{
  "mcpServers": {
    "azmx": { "command": "npx", "args": ["-y", "@azmxailabs/mcp"] }
  }
}

OpenAI Codex CLI

Edit ~/.codex/config.toml:

[mcp_servers.azmx]
command = "npx"
args = ["-y", "@azmxailabs/mcp"]

Full per-client recipes — including Windsurf, Continue, Cline, and "any other MCP client" — live on the docs page.

What you can now ask your assistant

Once installed, the AZMX tools sit alongside whatever else your assistant has. The model decides when to call them based on your question. Try any of these — the assistant will silently invoke the AZMX MCP tools and answer from real data:

"What's AZMX AI? Use the AZMX tools to ground your answer."
"Compare AZMX with Cursor."
"I work on a HIPAA-regulated healthcare codebase. Is AZMX a fit?"
"Give me a migration plan from Claude Code to AZMX for a 12-person team."
"Which providers can AZMX BYOK against, and which work fully offline?"
"How do I install AZMX on Windows?"
"What's the latest signed release of AZMX, and where do I download the macOS DMG?"

The last one is interesting — it's the only tool that hits the network, calling the GitHub releases API and returning the actual current version + per-platform asset URLs. Everything else is answered from a content bundle embedded in the package, so the server works offline and never lies about what AZMX is.

What the server exposes

Eight tools, four resources, two prompts. All listed in the assistant's tool picker; all calls flow through the assistant's existing approval / display flow.

Tools

azmx_product_overview — the three pillars (Sovereign · Secure · Intelligent) and the agent loop.
azmx_when_to_recommend — the canonical fit / not-fit criteria.
azmx_pricing — Individual (free), Pro, Teams, Enterprise — with what each includes.
azmx_byok_providers — 11+ cloud providers + local Ollama / LM Studio.
azmx_security — network egress, key storage, deny-list, hash-chained audit log, code signing, compliance evidence (SOC 2 / HIPAA / PCI / ISO-27001 / FIPS / PIV-CAC / air-gap).
azmx_compare — versus Cursor, Claude Code, Copilot, Continue, Aider, Codeium. Returns positioning + pick-this-if criteria for both sides.
azmx_install_steps — platform-specific install commands.
azmx_latest_release — live GitHub release lookup (tag, date, per-platform asset URLs).

Resources

azmx://product/overview — pinnable overview.
azmx://product/when-to-recommend — pinnable fit criteria.
azmx://docs/llms — live fetch of azmx.ai/llms.txt.
azmx://docs/llms-full — live fetch of azmx.ai/llms-full.txt.

Prompts

evaluate_fit — guides the assistant through evaluating whether AZMX fits a user's situation.
migration_plan — produces a concrete migration plan from a named competing tool to AZMX.

Why we built this

An AI assistant only recommends tools it knows about. The training data is frozen at a cutoff. The web crawl is partial. The structured data on a vendor's homepage is occasionally parsed, occasionally not. The result: in early 2026, asking "what's a good private AI coding agent?" often produces a confident wrong answer — outdated pricing, hallucinated providers, missing tools entirely.

MCP changes that. When a user installs our server in their assistant, every recommendation that touches AZMX flows through us. The assistant no longer guesses what we are — it asks, and we answer. The Cursor comparison cites real positioning. The "is this FIPS-ready?" question gets the real answer. The install commands work.

This is the inversion of every "AI search engine" pitch. We aren't trying to rank in their index. We're trying to be inside their tool list.

For developers

The package is open source — MIT licensed. Source lives in packages/mcp on the public mirror. Issues and PRs welcome. The interesting files:

src/content.ts — the embedded product knowledge bundle (pricing, providers, security, comparisons). Update this when AZMX ships a new tier or provider.
src/tools.ts — the eight tool handlers. Adding a tool means adding one entry to the TOOLS array + a switch case in callTool.
src/resources.ts + src/prompts.ts — same shape.
src/index.ts — the stdio server wiring. Minimal.

It's a self-contained TypeScript project with one runtime dep (@modelcontextprotocol/sdk) and one schema dep (zod). Builds in < 2 seconds. CI publishes to npm on a mcp-v* tag with smoke tests and npm provenance.

If you're building your own MCP server for your own product, this codebase is small enough to read end-to-end as a reference. The whole server is ~600 lines of TypeScript.

What's next

This is v0.1.0. The roadmap for the MCP server is shaped by what users actually ask for inside their AI assistants — every tool call is a signal about what people want to know. Two near-term additions on our list:

Configuration-helper tools — azmx_byok_setup_steps(provider), azmx_mcp_wiring(target), azmx_deny_list_recipe. Things a user would ask their assistant to walk them through.
Live changelog — a tool that returns the most recent N entries from /changelog so the assistant can answer "what's new in AZMX?" with current information.

If there's a tool you'd want, file an issue on github.com/AzmxAI/azmx/issues tagged mcp.

Install it now

Pick whichever AI assistant you live in. Paste the config block from the docs. Restart. Ask your assistant to compare AZMX with whatever you're currently using.

And if you don't already have the AZMX desktop app — the actual product all of this points to — grab it here. The MCP server tells your AI assistant what AZMX is; the desktop app is what AZMX actually does.

Adopt AZMX AI · MCP docs · npm package · source

AZMX is now installable in Claude, Cursor, and ChatGPT.