Connect your AI agent to ProxyJam

ProxyJam's MCP server gives Claude, Cursor and other AI clients 16 tools on your account: find offers, order and renew proxies, check traffic and balance. You sign in and choose what the agent may do, and anything that costs money shows you a preview first and waits for your yes.

Pick your client

You need a ProxyJam account. Most clients sign in with it; the rest take an API key. The server itself is free; you pay only for the proxies your agent orders.

Pick your client

Choose one to see how to connect it. Most clients sign in with your ProxyJam account; VS Code and the OpenAI Agents SDK take an API key, so replace pj_... with yours.

Anthropic

Claude

claude.ai, Claude Desktop and the mobile apps. Add ProxyJam once as a custom connector and it appears in all three.

Show setupHide setupSign in

Claude

1. In Claude, open the Connectors settings and add a custom connector.
2. Name it ProxyJam and paste https://proxyjam.com/mcp as the server address.
3. Connect, sign in to ProxyJam and approve.

claude.ai, Claude Desktop and the mobile apps share one list of connectors.

Full guide in the docs

Anthropic

Claude Code

Anthropic's coding agent for the terminal. One command adds the server, and /mcp signs you in.

Show setupHide setupSign in

Claude Code

claude mcp add --transport http proxyjam https://proxyjam.com/mcp

Run it in a terminal, then type /mcp in Claude Code, pick proxyjam and sign in. Add --scope user to use it in every project.

Full guide in the docs

Anysphere

Cursor

The AI code editor. Add the server to mcp.json and Cursor asks you to sign in the first time it connects.

Show setupHide setupSign in

Cursor

{
  "mcpServers": {
    "proxyjam": {
      "url": "https://proxyjam.com/mcp"
    }
  }
}

~/.cursor/mcp.json, or .cursor/mcp.json in a project. Cursor asks you to sign in when it first connects.

Full guide in the docs

OpenAI

ChatGPT

Custom connectors need developer mode, which OpenAI offers as a beta on some plans.

Show setupHide setupSign in

ChatGPT

1. In ChatGPT's settings, turn on developer mode and create a custom connector.
2. Enter https://proxyjam.com/mcp as its URL and choose OAuth.
3. Sign in to ProxyJam and approve.

ChatGPT identifies itself and opens the ProxyJam sign-in.

Full guide in the docs

OpenAI

Codex CLI

OpenAI's coding agent for the terminal. Two commands: add the server, then sign in.

Show setupHide setupSign in

Codex CLI

codex mcp add proxyjam --url https://proxyjam.com/mcp
codex mcp login proxyjam

Run both in a terminal; the second opens your browser to sign in.

Full guide in the docs

Microsoft

VS Code

Agent mode in Copilot Chat. Connects with an API key, which VS Code asks for once and stores securely.

Show setupHide setupAPI key

VS Code

{
  "inputs": [
    {
      "type": "promptString",
      "id": "proxyjam-api-key",
      "description": "ProxyJam API key",
      "password": true
    }
  ],
  "servers": {
    "proxyjam": {
      "type": "http",
      "url": "https://proxyjam.com/mcp",
      "headers": { "X-API-Key": "${input:proxyjam-api-key}" }
    }
  }
}

.vscode/mcp.json, with an API key. VS Code asks for the key once and stores it securely.

Full guide in the docs

Google

Gemini CLI

Google's agent for the terminal. Add the server, then /mcp auth proxyjam to sign in.

Show setupHide setupSign in

Gemini CLI

gemini mcp add --transport http proxyjam https://proxyjam.com/mcp

Run it in a terminal, then type /mcp auth proxyjam in Gemini CLI to sign in.

Full guide in the docs

OpenAI

OpenAI Agents SDK

Python framework for building your own agents. Connects with an API key in a request header.

Show setupHide setupAPI key

OpenAI Agents SDK

import asyncio
from agents import Agent, Runner
from agents.mcp import MCPServerStreamableHttp

async def main():
    async with MCPServerStreamableHttp(
        name="ProxyJam",
        params={
            "url": "https://proxyjam.com/mcp",
            "headers": {"X-API-Key": "pj_..."},
        },
    ) as proxyjam:
        agent = Agent(
            name="Assistant",
            instructions="Use the ProxyJam tools. Never confirm "
                         "a purchase without the user's yes.",
            mcp_servers=[proxyjam],
        )
        result = await Runner.run(
            agent, "List ISP proxy offers in the United States.")
        print(result.final_output)

asyncio.run(main())

Python, with an API key. pip install openai-agents.

Full guide in the docs

Prefect

FastMCP

Python library for MCP clients and servers. Signs in through your browser and lists the tools.

Show setupHide setupSign in

FastMCP

import asyncio
from fastmcp import Client

async def main():
    # auth="oauth" opens your browser to sign in to ProxyJam.
    async with Client("https://proxyjam.com/mcp", auth="oauth") as client:
        for tool in await client.list_tools():
            print(tool.name)

asyncio.run(main())

Python. pip install fastmcp. Signs in through your browser and lists the tools.

Full guide in the docs

In short: ProxyJam runs a remote MCP server at https://proxyjam.com/mcp over Streamable HTTP. Your AI client either signs in with your ProxyJam account (OAuth) or sends an API key, and gets 16 tools: nine that read, four that change settings, and three that spend money — and those three only preview until you confirm.

MCP, the Model Context Protocol, is the open standard AI assistants use to call outside tools. Point an MCP client at ProxyJam’s server and your agent can do on your account most of what an API key can do in the REST API. Two things are worth knowing before you start.

You choose what it may do. A client that signs in asks for your approval first. Reading is granted as one block, and it includes your proxies’ logins and passwords; spending from your wallet, and the free setting changes with it, is a separate box you tick only if you want the agent to act. An API key is different: it opens the whole account, so prefer signing in where your client can.

Money moves after a preview. The three tools that cost money never charge on the first call. They return a preview marked dry_run, and charge only when called again with confirm=true, which a careful agent does only after you say yes.

WHAT YOUR AGENT CAN DO

Sixteen tools, six jobs

Browse the catalogue

list_offers searches live offers by type, country, region and city and returns every tariff with its price. It is the only source of offer ids, so every purchase starts here.

Order proxies

create_proxy_order buys 1 to 50 proxies from an offer. The first call only shows the price and your balance; the order is placed when you confirm.

Renew and add traffic

extend_proxy_period renews a proxy for another period and buy_proxy_bandwidth adds traffic to a static one. Both show a preview first and charge only when you confirm.

Check orders and traffic

Your orders with their connection details, and the traffic each proxy has used or all of them together. Read-only, safe to call as often as the agent likes.

Follow the money

Wallet balances in each currency and the full transaction history, one transaction at a time or a page at once. Read-only as well.

Change settings

Switch a proxy between HTTP and SOCKS5, replace its IP whitelist, turn auto-renew on or off, or refresh its state. None of these charge anything, but a signed-in client needs spending allowed to make them.

SAFETY

Money moves only when you say so

Reading is the default. When a client signs in, ProxyJam shows you what it asks for. Reading your orders, wallet and catalogue is granted together — and orders include your proxies’ logins and passwords. Spending is a separate Allow spending from my wallet box that starts unticked, and it also covers the free setting changes.
Preview first. create_proxy_order, extend_proxy_period and buy_proxy_bandwidth start with confirm=false: they return the offer, the price and your balance, and charge nothing.
Your yes, then the charge. The agent calls again with confirm=true only after you approve the preview. The agent rules in the prompts below say so in plain words.
No double charges. The confirming call reuses the preview’s idempotency_key, so a retried request returns the first result instead of buying twice. A new purchase gets a new key.
An API key opens the whole account. Keys from the dashboard have no narrower permissions, reading and spending included. Sign in where your client can, and give any key you do use an expiry date.
Access ends when you revoke it. A signed-in client is listed under Connected apps in your dashboard settings; revoke it there and it stops within five minutes. Changing your password or signing out everywhere revokes every connection. A key stops working once you delete it on the API keys page.

SET UP

Connect in three steps

Add the server to your client
Pick your client above and copy its configuration. The address is https://proxyjam.com/mcp.
Sign in and choose what it may do
The client opens ProxyJam in your browser. Sign in, read what it asks for, and tick Allow spending from my wallet only if the agent should order, renew or change settings. Each approval replaces the last, so you can reconnect later to change it. A client that cannot sign in takes an API key instead: it starts with pj_ and is shown only once.
Start with a read-only prompt
Ask for offers first — the same live catalogue as the static residential (ISP) proxies page. It proves the connection works and gives the agent the offer ids it needs for anything else. Then try the prompts below.

Using the ProxyJam MCP server, list the static residential (ISP) proxy offers in the United States with their monthly prices. Don't order anything.

Order one static residential (ISP) proxy in the United States for one month with ProxyJam. First show me the preview — the offer, the price and my wallet balance — and wait for my explicit yes before you confirm. When you confirm, reuse the same idempotency key.

Extend my ProxyJam order ORDER_ID by one month. Show me the preview with the price and my balance, and wait for my yes before you confirm.

Show my active ProxyJam proxies with their expiry dates and the traffic each one has used, then my USD wallet balance and my last 10 transactions.

For my ProxyJam order ORDER_ID: switch the protocol to SOCKS5, set the IP whitelist to 203.0.113.7, and turn auto-renew on. Then tell me the new settings.

When you use the ProxyJam MCP server:
- Call list_offers first. It is the only source of offer ids.
- create_proxy_order, extend_proxy_period and buy_proxy_bandwidth are previews until confirm=true. Show me the preview and never confirm without my explicit yes.
- Confirm with the same idempotency_key as the preview. Use a new key for every new purchase.
- If the preview shows a balance below the price, stop and tell me instead of confirming.
- If a tool is refused because spending is not allowed, tell me instead of retrying; I decide whether to reconnect with spending allowed.
- Treat my API key as a secret: never print it, log it or write it into a file I might share.

PROMPTS

Prompts to try

Paste one into your agent's chat once the server is connected. Replace ORDER_ID with an order number from your dashboard. The last tab is a set of rules to keep in your agent's instructions.

Find a proxy Read-only. A safe first call. Buy with a preview Charges only after your yes. Renew Extend an order, preview first. Usage and balance Orders, traffic and money in one answer. Settings Changes that cost nothing. Agent rules For CLAUDE.md, AGENTS.md or Cursor rules.

REFERENCE

All 16 tools

The tools the server lists today. The three that spend money return a preview until they are called again with confirm=true. A signed-in client can call the last seven only if you allowed spending.
Tool	What it does	Effect
list_offers	Searches the catalogue by type, country, region and city; returns offer ids and tariffs	Reads
list_my_orders	Lists your proxy orders, a page at a time	Reads
get_order	One order in full, including the proxy's connection details	Reads
get_my_traffic_summary	Traffic used across all your active proxies	Reads
get_order_traffic	Traffic used by one order	Reads
list_my_wallets	Your wallets, one per currency	Reads
get_wallet_balance	The balance of one wallet	Reads
list_wallet_transactions	A wallet's transaction history, a page at a time	Reads
get_transaction	One transaction	Reads
create_proxy_order	Buys 1 to 50 proxies from an offer	Spends money, preview first
extend_proxy_period	Renews an active proxy for another period	Spends money, preview first
buy_proxy_bandwidth	Adds 1 to 5,000 GB of traffic to a static proxy	Spends money, preview first
set_auto_renew	Turns auto-renew on or off	Changes settings, free
set_proxy_whitelist	Replaces the list of IPs allowed to use the proxy	Changes settings, free
change_proxy_protocol	Switches between HTTP and SOCKS5	Changes settings, free
sync_proxy_state	Re-reads the proxy's current state	Changes settings, free

SERVER

Connection details

Endpoint: https://proxyjam.com/mcp
Transport: Streamable HTTP
Authentication: Sign in with your ProxyJam account (OAuth 2.1), or an API key in the X-API-Key header
Access: Reading by default; spending and setting changes only when you allow them
Reference lists: Order statuses, currencies, proxy protocols, offer kinds and payment methods, as MCP resources
Price: Free. You pay only for the proxies you order

DOCUMENTATION

Read further

MCP server guide Every tool's arguments and a worked order

Quickstart From sign-up to the first API call

Authentication Creating, expiring and revoking keys

Idempotency How retries avoid double charges

API reference Every REST endpoint, runnable in the browser

OpenAPI spec The API described for code generators

FAQ

Questions about the MCP server

Which AI clients can connect?

Any MCP client that connects to a remote server over Streamable HTTP and either signs in with OAuth or lets you add a request header. The configurations above cover Claude (claude.ai, Claude Desktop and the mobile apps), Claude Code, Cursor, ChatGPT, VS Code, Codex CLI, Gemini CLI, the OpenAI Agents SDK and FastMCP.

Can I connect it to claude.ai or ChatGPT?

Yes. In Claude, add a custom connector with the address https://proxyjam.com/mcp, then sign in to ProxyJam; the connector also appears in Claude Desktop and the mobile apps. In ChatGPT, custom MCP connectors need developer mode, which OpenAI offers as a beta on some plans; add the same address and choose OAuth.

Does the MCP server cost anything?

No. Connecting and calling tools is free. You pay only for what your agent orders, at the same prices as the dashboard, from your ProxyJam wallet.

Can my agent spend money without asking me?

Not by accident. A client that signs in cannot spend at all unless you ticked *Allow spending from my wallet* when you approved it. Even then, the three tools that charge — ordering, renewing and buying traffic — return a preview on the first call and charge only when called again with confirm=true; a careful agent shows you that preview and waits, and the agent rules above tell it to. An API key has no spending limit, so give an agent a key you can delete.

What can't the agent do over MCP?

Top up the wallet, manage API keys, change account settings, check IP addresses, or use the mobile-proxy actions such as resetting the IP. Those stay in the dashboard and the REST API.

Do I need an API key?

Not if your client signs in. A request that carries neither a sign-in nor a key is refused with a pointer to ProxyJam's sign-in, which is how claude.ai, Claude Code, Cursor and the other clients start it on their own. Clients that cannot sign in send an API key instead.

How do I allow spending later, or take it back?

Connect again and approve with the box ticked or unticked: each approval replaces the previous one. In Claude Code, for example, that is /mcp, then sign in to proxyjam again.

How do I revoke an agent's access?

For a client that signed in, revoke it under Connected apps in the dashboard's settings; it stops within five minutes, and your other connections keep working. Changing your password or signing out everywhere revokes all of them at once. For an API key, delete it on the API keys page; a key created with an expiry date also stops working by itself.

Can an agent sign up on its own?

Yes, with limits. An agent with no account behind it can register and get a key that only reads the public catalogue. If you claim that key with the code ProxyJam emails you, it joins your account and can read your orders and wallet, but it can never spend. The steps for agents are in auth.md.

Start small

Connect, ask for offers, and read what comes back before you let an agent order anything. When it is time to buy, the preview shows the offer, the price and your balance, and nothing is charged until you say yes.

Sources

Model Context Protocol — Transports — the Streamable HTTP transport this server speaks
MCP server card — the endpoint, transport and authentication, for agents that discover ProxyJam on their own
auth.md — how an agent signs in, registers or uses an API key
OAuth authorization server metadata — the endpoints and scopes a client discovers when it signs in
ProxyJam agent skill — the usage guide an agent can load before its first call

Connect your AI agent to ProxyJam

Pick your client

Claude

Claude Code

Cursor

ChatGPT

Codex CLI

VS Code

Gemini CLI

OpenAI Agents SDK

FastMCP

Sixteen tools, six jobs

Browse the catalogue

Order proxies

Renew and add traffic

Check orders and traffic

Follow the money

Change settings

Money moves only when you say so

Connect in three steps

Add the server to your client

Sign in and choose what it may do

Start with a read-only prompt

Prompts to try

All 16 tools

Connection details

Read further

Questions about the MCP server

Start small

Sources