Beacon: The BitVector MCP Server

Beacon is the bridge between AI agents and the BitVector ecosystem — it lets your LLM-powered assistants browse models, discover workflows, and submit render requests, all through a secure, human-in-the-loop pipeline.

What is Beacon?

Beacon is a Model Context Protocol (MCP) server that gives AI agents structured access to BitVector's rendering capabilities. Instead of your agent guessing at prompts or hallucinating model names, Beacon provides:

• Live model & workflow discovery — Your agent can browse available renderers, check their capabilities, and pick the best one for your request
• Community skill integration — Access to community-made prompt guides and workflows
• Secure render submission — Every render request goes through BitVector's approval queue, and must be approved by you, before touching a GPU upstream

Experimental & Advanced

Beacon is not only an experimental project, but also has a more advanced target audience. Some edges may still be sharp!

You'll need either pre-existing knowledge of "What is an MCP server and how do I use one?", or a willingness to learn, in order to use Beacon. Providing instructions on how to setup your own Agent and/or MCP servers is a bit out of our scope of support.

Beacon is by no means intended to replace our normal ecosystem surfaces (Vector on Discord, and Spyglass on the web) which are intended for users of any experience level.

Think about Beacon as being a lighthouse for Agents, showing them how to navigate through BitVector & Graydient along with you!

Psst... that is actually the idea behind the name "Beacon".

Why? What about Skills?

Graydient Skills are a fantastic way to easily go from idea -> result, however what if the result wasn't quite what you wanted?

The inspiration behind Beacon is the concept of

"What if we took the conversational aspect of LLMs and Agents, which is what they're built for, and gave them the ability to put them in the driver's seat?"

Beacon lets your Agent be your painter, Beacon's tools act as your Agent's hands, Graydient acts as the paintbrush, and then the vast collection of models & workflows act as the canvas!

Beacon also lets your Agent access Graydient Skills too, so the collection of community built guidelines/templates are still useful to your Agent too. Skills work in tandem with Beacon and your Agent!

You and your Agent can build an idea together, execute it, and then continue to iterate upon it until it's perfect. The final evolution of "closing the creative loop" in a way.

Humans (In The Loop) Required!

This is the core philosophy behind Beacon: AI agents propose, humans approve. Beacon intentionally has no "auto-approve" mode. Here's why:

1. Render costs are real — Every generation uses paid Graydient GPU time
2. Creative intent matters — You know what you want better than any agent does
3. Safety by design — No autonomous NSFW (or potentially malicious, even if accidental) generation without explicit human consent
   • Note: You are intended to actually review what your Agent wants to render. As the person who signs the official approval, you are responsible for any renders generated and their contents! Please ensure that these still follow both BitVector and Graydient's Terms of Service!

When your agent submits a render, you'll be prompted to approve it in one of two ways:

• Spyglass (web app) — If you have it open, the approval appears there instantly on the approvals page
• Vector (Discord bot) — If Spyglass isn't open, Vector sends you an interactive DM with embed details and approval buttons. You will of course need a Discord account connected to your BitVector account for this option to work.

Smart Routing

Navigator automatically detects whether you have Spyglass open and routes accordingly — so you won't get spammed with Discord DMs when you're already on the web app!

Your agent will wait up to 25 seconds for you to approve. If you don't approve within that time frame, your agent will nudge you to approve the request. After 5 minutes, the request is automatically rejected and your agent will need to resubmit.

Demo

Here's a clip of Beacon in action! Simply ask your Agent something to render (even if vague!), review what it proposes, and kick back as it handles the rest!

TIP

This clip is of course sped up, process times depends on the speed of your LLM, which workflow you (or it) chooses to use, etc. No sense in you waiting here though for too long!

Quick Start

If you've already got an LLM client or agent harness that supports MCP, getting started is a single command:

npx --allow-git=all --yes github:BitVectorApp/Beacon#v1 auth

Or, if you're using a headless system (such as managing over SSH), you can use the manual / headless authentication flow then follow the instructions provided:

npx --allow-git=all --yes github:BitVectorApp/Beacon#v1 auth --manual

On first run, Beacon will open your browser for BitVector authentication. After that, plug Beacon into your client/harness' MCP server list, you're ready to go — tell your agent "generate an image of..." and let it handle the rest.

Need a client?

Beacon is an MCP server — you'll need an MCP-compatible client to use it. If you're already using an AI agent platform or coding assistant, check if it supports MCP servers. The Model Context Protocol site has a list of compatible clients.

You still need to bring your own LLM and client (or a full blown Agent!), but we do provide you with the bridge to integrate your LLM/Agent with BitVector & Graydient.

If your LLM/Agent has internet access, point it to the Beacon GitHub Repository and ask it for assistance with getting Beacon up and running!

How It Works

flowchart LR Agent[Your AI Agent] -->|MCP| Beacon Beacon -->|OAuth mcp scope| Navigator Navigator -->|Approval Queue| You[You!] You -->|Approve| Navigator Navigator -->|Submit| Graydient[Graydient GPU] Graydient -->|Render| Navigator Navigator -->|Result URL| Beacon Beacon -->|Image/Video| Agent

1. Your agent asks Beacon to generate something
2. Beacon sends the request through to Navigator (BitVector's backend)
3. Navigator pauses the request and asks you to approve it (via Spyglass or Vector)
4. Once you approve, Navigator sends it to Graydient for rendering
5. The result flows back through the chain to your agent

Release Channels

Channel	Command	Description
v1	github:BitVectorApp/Beacon#v1	Stable — recommended for everyday use
dev	github:BitVectorApp/Beacon#dev	Latest changes, may have sharp edges

Environment Variables

Beacon is configured through environment variables. The most common ones:

Variable	Default	Description
MCP_TRANSPORT	stdio	Transport mode: stdio or http
MCP_HOST	127.0.0.1	Host for HTTP transport mode
MCP_PORT	3333	Port for HTTP transport mode
BEACON_ALLOW_ADULT_METADATA	0	Set to 1 to expose community-flagged adult metadata

NSFW Content

Graydient is an uncensored platform, and some model/workflow/skill metadata may contain adult-themed language. By default, Beacon attempts to filter community-flagged NSFW metadata. If you enable BEACON_ALLOW_ADULT_METADATA, be aware that your inference provider may have policies about this kind of content. Check their Terms of Service first.

As NSFW metadata is community-flagged rather than on an authoritative source level, we cannot make any guarantees that it will all be filtered out even when not opted-in.

Going Deeper

For complete technical documentation — including manual authentication for headless environments, HTTP transport setup, cache behavior, and contributing — head over to the Beacon GitHub Repository.

Beacon is open source (AGPL-3.0) and we welcome contributions! Just open an issue first to make sure your idea aligns with Beacon's direction.