Question 1

What is the LightningBytes MCP server?

Accepted Answer

It is a Model Context Protocol server served as a Streamable HTTP route at /api/mcp. An MCP client such as Claude Code, Claude Desktop, Cursor or VS Code connects to that one endpoint and gets a list of tools it can call to manage the account's proxies: products and locations, sub-users, allocations, endpoint generation and rotation, proxy checks and usage reads.

Question 2

Which clients does it work with?

Accepted Answer

Any Streamable HTTP MCP client. The dashboard ships copy-ready config blocks for Claude Code, Claude Desktop, Cursor, VS Code and a generic client. Clients differ in how they name the transport, so each block carries the field that client needs: Roo Code and Zoo Code require a type of streamable-http, VS Code uses type http, and Cursor infers it from the url.

Question 3

How do I authenticate?

Accepted Answer

With an API token created on the dashboard's API Tokens tab, sent as a bearer credential in the Authorization header. The value is "Bearer <token>", and the space matters: a token pasted without the Bearer prefix fails with a 401. Tokens are shown once when created, so the config blocks use a placeholder you replace yourself.

Question 4

Does the token's scope still apply?

Accepted Answer

Yes. An API token carries a set of product and pool pairs, and MCP enforces exactly the same pairs as the HTTP API. A call outside that scope is refused before it reaches the operation, so an agent can only act where the token was granted.

Question 5

Why does the server not use OAuth?

Accepted Answer

OAuth 2.1 is the right fit when a third party must act on a human's behalf without holding a long-lived credential. This server is the machine-to-machine case: the client is a token the account owner created, scoped and can revoke at any time, which is the pre-registered client pattern the protocol keeps for non-interactive callers.

Question 6

Can MCP tools buy traffic or read invoices?

Accepted Answer

No. The server is operational only. No tool buys traffic, creates a checkout, changes a plan or reads an invoice, and the MCP layer imports no payment code at all. The single buy affordance, a top-up link tool, returns a dashboard URL for a human to open and complete.

Question 7

What can an agent actually do through the tools?

Accepted Answer

Read the token's identity and scope, list products, pools and covered locations, then manage sub-users (list, create, delete, reset a proxy password, enable or disable, set an allocation), generate endpoints and rotate their exit IPs, check an endpoint's connectivity and exit IP, and read balances, traffic totals, history and breakdowns.

Question 8

Does an agent see other accounts' usage?

Accepted Answer

No. Usage reads are scoped to the account that owns the token, filtered on that account's own upstream usernames. The underlying supplier statistics are reseller-wide, so an unfiltered read would expose other customers; the tools never forward a client-supplied username and always resolve the account's own users first.

Question 9

Is the server stateful?

Accepted Answer

No. It runs stateless: a fresh server instance is built per request and closed over that request's authentication, no session id is issued and responses are plain JSON. That means there is nothing to share between requests or tenants, and the endpoint scales behind an ordinary load balancer.

Question 10

Which URL should go in a client config?

Accepted Answer

Use the www host: https://www.lightningbytes.com/api/mcp. The apex host redirects to www, and an HTTP client drops the Authorization header when a redirect changes the host, so an apex URL fails with a 401 on a valid token. Use the exact URL from the dashboard's MCP tab.

Question 11

Does the client need to poll for changes?

Accepted Answer

No. A client asks the server for its tool list when it connects, so the tools appear in the client's list immediately. Some clients also open the transport's optional server-to-client stream, which this endpoint serves as an authenticated event stream rather than refusing it.

Question 12

How is this different from the Developer API?

Accepted Answer

The Developer API is the HTTP surface your code calls directly. The MCP server is the same operations packaged as tools a language model can discover and call, so an agent drives the account without writing HTTP requests. Both call the same operations layer, so behaviour cannot drift between them.

Give your agent the account, not a pile of curl calls

Connect the client you already use

One endpoint, typed tools

Everything that powers mcp server

Streamable HTTP

The tokens you already have

Scope enforced

Twenty-one tools

Sub-user management

Endpoints and rotation

Connectivity checks

Usage reads

Same code as the API

Operational only

Configs for real clients

Manual and guide

Why the MCP server matters

Fewer turns, fewer errors

No new credential

The whole lifecycle

Where this feature does the work

Who works on LightningBytes

Web scraping

Multi-account management

AI agents

Social media automation

Affiliate marketing

Ad verification

E-commerce monitoring

SERP tracking

Plug into the tools you already use

Questions, answered

Start working with cleaner IPs