How to Connect Your Existing Agent to ATXP
Most guides for adding capabilities to an AI agent follow the same pattern: create an account, copy an API key, paste it into a config file, update your system prompt, restart, debug. It’s the kind of setup friction that makes sense when software is installed by humans, for humans.
ATXP works differently. Your agent sets itself up.
The instruction is the setup
Paste this into your agent chat — or just type it:
npx atxpThat’s the complete instruction. Your agent runs the command, which creates an ATXP account (or links an existing one), generates and stores an API key, and configures itself with the tools it needs. You don’t touch a config file. You don’t copy anything. The agent handles its own onboarding.
This works because agents that can run shell commands can bootstrap their own infrastructure. Most capable agents today — Claude Code, MANUS, custom agents with tool use enabled — can execute npx atxp the same way they’d run any other command you ask them to. The setup that would take a human five minutes of back-and-forth happens in seconds, inside the same conversation where you’re already working.
What your agent gets
Once the command runs, your agent has four things it didn’t have before:
Identity. A unique ATXP account with verifiable credentials. Your agent can prove who it is to any service that checks.
Payments. Pay-per-use billing through the ATXP gateway. Your agent can transact on the internet — paying for API calls, tools, services — without you setting up billing separately for each one.
Email. An @atxp.email address. Your agent can send and receive email as itself, not as you.
Tools. LLM inference, web browsing, search, image generation, code execution, and more — available to your agent as callable tools, with per-use pricing built in.
Free to create. Your agent only pays for what it uses after that.
Why this matters
The standard model for giving an agent capabilities is human-mediated: you configure the capability, then hand it to the agent. ATXP inverts this. The agent requests the capability itself, and the setup happens autonomously.
This is meaningful for a few reasons.
First, it removes you as a bottleneck. You don’t need to anticipate every capability your agent will need before it needs it. If your agent is working on something and realizes it needs to send an email or make a payment, it can provision that capability on demand.
Second, it’s the right mental model for where agents are going. Agents are increasingly running as autonomous processes — not as chat sessions you’re watching, but as background workers acting on your behalf. Those agents need their own infrastructure: their own identity, their own payment method, their own email. ATXP is that infrastructure, and npx atxp is how an agent claims it.
If you’re building an agent
You don’t need to pre-configure ATXP for your users. Your users’ agents can self-provision when they need it.
If you want ATXP available from the start, you can include npx @atxp/mcp-server in your agent’s initialization — it installs the MCP server and makes the full ATXP toolset available immediately. But you don’t have to. The on-demand path works fine.
Either way, your users’ agents get an ATXP account that belongs to them, billed to them, with their own credentials. You stay out of the loop.
Agent self-provisioning is the process by which an AI agent creates and configures its own infrastructure — identity, payment credentials, and communication channels — without human intervention. Rather than a developer manually setting up accounts and copying API keys, the agent executes a setup command and bootstraps its own operational environment.
npx atxpGive your agent identity, payments, email, and tools in one command — no config files required. What is an agent account? → · How agents send email →
If something doesn’t work, email support@atxp.ai. Full API reference at docs.atxp.ai.
Frequently asked questions
What does npx atxp actually do?
It creates an ATXP account for your agent (or links an existing one), generates and stores an API key, and configures the agent with identity, payments, email, and tools — all in a single command your agent runs itself.
Does my agent need special permissions to run npx atxp?
Any agent that can execute shell commands can run npx atxp. Most capable agents today — Claude Code, MANUS, custom agents with tool use enabled — support this natively.
What capabilities does my agent get after connecting?
Identity (verifiable credentials), payments (pay-per-use billing through the ATXP gateway), email (an @atxp.email address), and tools including LLM inference, web browsing, search, image generation, and code execution.
Does connecting ATXP cost anything upfront?
Creating an ATXP account is free. Your agent only pays for what it actually uses — each tool call, email sent, or service accessed is billed individually.
Can I pre-configure ATXP for my users instead of having each agent self-provision?
Yes. You can include npx @atxp/mcp-server in your agent’s initialization to install the MCP server and make the full ATXP toolset available from the start. But the on-demand path via npx atxp also works fine.
What happens if my agent already has an ATXP account?
Running npx atxp detects the existing account and links to it rather than creating a duplicate. The command is safe to run multiple times.