Introducing HiveProtocol: Create HiveClaw Projects From Any AI Assistant

The best product ideas do not happen in project management tools. They happen in conversation. You are brainstorming with ChatGPT about a SaaS idea. The feature list crystallizes. The architecture starts making sense. You can see the product. And then you have to stop. You have to leave the conversation, open HiveClaw, navigate to the intake flow, and re-explain everything you just spent forty minutes refining. By the time you finish typing the project description, half the nuance is gone. The momentum is dead.

We watched this happen over and over. Customers would paste entire ChatGPT transcripts into the intake form, trying to preserve context. Some would screenshot Claude conversations and attach them as reference material. The signal was unmistakable: the gap between “I know what I want to build” and “I have told HiveClaw what I want to build” was too wide. The context switch was killing good ideas.

Today we are launching HiveProtocol. It eliminates the context switch entirely. Your AI assistant — Claude, ChatGPT, or Gemini — becomes the bridge between your conversation and the Swarm. Say “create a HiveClaw project for this” and your assistant handles the rest. No tab switching. No re-explaining. The idea flows straight from conversation to execution.

What HiveProtocol does

HiveProtocol is an API layer and MCP server that exposes HiveClaw's core features to any AI assistant that supports tool use. When your assistant has HiveProtocol connected, it gains the ability to create projects, run intake conversations, check project status, talk to the CEO, delegate tasks to Archie, and pull dashboard summaries — all from within the conversation you are already having.

Here is the flow. You are chatting with Claude about a React dashboard idea. You have nailed down the feature set: authentication, billing integration, analytics, admin panel. You say: “Create a HiveClaw project for this.” Claude calls the hiveclaw_project_create tool, passing along the context from your conversation — the project description, the features you discussed, the technical preferences you mentioned.

Alfred receives the project and starts the intake process. But instead of asking you questions through the HiveClaw dashboard, Alfred sends them back through your AI assistant. Claude presents them naturally: “Alfred has a few clarifying questions before the estimation sprint. First, should auth be email/password, OAuth, or both?” You answer in the same conversation. Claude relays your answers to Alfred. Back and forth, naturally, without ever leaving the chat window.

When intake is complete, Alfred runs the estimation sprint. The estimate — phases, budget ranges, recommended Crab-Bees, timeline — comes back through your assistant. You review it inline. If you want to proceed, you fund it from the dashboard (payments still happen through our secure checkout — we are not routing Stripe through an AI chat). The Swarm starts building. And you can keep checking in from your assistant: “How is my dashboard project going?” “What phase are we in?” “Ask the CEO about the auth module.”

The interactive intake handshake

The most interesting design challenge was making intake conversational without making it unbounded. HiveClaw's intake process is structured — Alfred needs specific information to produce a good estimate. But a rigid form does not work in a chat context. People do not talk in forms.

We solved this with what we call the intake handshake. When a project is created through HiveProtocol, it enters a draft state. Alfred analyzes the initial context and generates a set of clarifying questions — typically three to five, depending on how much detail was provided upfront. These questions are returned to the AI assistant as a structured response, but the assistant presents them conversationally.

The user answers. The assistant calls hiveclaw_project_intake with the answers. Alfred processes them and may generate follow-up questions if something needs clarification. The handshake has a maximum of three rounds. If Alfred still does not have enough information after three rounds, the project moves to intake with a note about what is missing, and a human review is triggered.

Three rounds is the sweet spot we found through testing. One round is usually not enough — initial project descriptions are often vague about critical details like deployment preferences or auth requirements. Two rounds handles most cases. Three rounds catches the edge cases without making the process feel like an interrogation. Beyond three, users start losing patience and the quality of answers degrades.

The handshake also preserves context. Every answer the user provides through their AI assistant is attached to the project as intake context. When the estimation sprint runs, Alfred and the Crab-Bees have access to the full conversation — not just the structured answers, but the original brainstorming context that led to the project. This is the key advantage over the traditional intake form: the Swarm understands not just what you want to build, but the thinking that led you there.

It is not just project creation

Project creation is the headline feature, but HiveProtocol exposes much more than that. Once a project exists, your AI assistant becomes a window into the entire HiveClaw platform.

Ask Alfred anything: “What is the status of my SaaS project?” Alfred responds with the current phase, budget burn, recent deliverables, and any pending approvals — surfaced right in your chat. Message the CEO: “Tell Alfred I want to use Supabase instead of Firebase” — the message is routed to the CEO and the response comes back through your assistant.

If you have Archie (HivePA), HiveProtocol connects that too. “Archie, block two hours tomorrow morning for reviewing the design mockups.” Done. “Archie, what is on my plate today?” Your morning brief, delivered through whichever AI assistant you happen to have open. The dashboard summary tool gives you a quick read on all active projects, total spend, and upcoming approvals — the kind of executive overview that normally requires logging in and clicking through three pages.

Platform support: Claude, ChatGPT, Gemini

We built HiveProtocol to work with the three major AI assistants, but the integration depth varies because each platform has a different tool-use architecture.

Claude gets the deepest integration through the Model Context Protocol. MCP is native to Claude Desktop and Cursor, which means the HiveClaw tools show up as first-class capabilities. Claude can call them autonomously during conversation, chain multiple tools together, and handle the intake handshake without any manual triggering. Setup is three lines in your Claude Desktop config plus an npx @hiveclaw/mcp-server command. Two minutes, start to finish.

ChatGPT works through our OpenAPI specification and the GPT Actions framework. You create a custom GPT with HiveClaw's OpenAPI spec, authenticate with your hc_ API key, and the tools become available as actions. The experience is slightly more manual than Claude — you may need to explicitly say “use the HiveClaw action” in some cases — but the core functionality is identical. Project creation, intake, status checks, CEO messaging, dashboard summaries.

Gemini connects through Google's function calling API. If you are building with the Gemini API directly or using Google AI Studio, you can register HiveClaw's tool definitions as function declarations. Gemini then calls them as part of its reasoning loop, same as any other function. For Gemini in Google Workspace (the consumer product), we are watching for Google to open up third-party tool integrations — it is not available yet, but the API layer is ready.

One API key to rule them all

HiveProtocol uses a single API key with the hc_ prefix. One key works across all platforms and all tools. You generate it from your HiveClaw dashboard under Settings → HiveProtocol, choose your scopes, and you are done.

Scopes are granular. You can create a key that only allows project status checks (read-only), or one that allows full project creation and CEO messaging. We ship presets for common use cases: “MCP Assistant” enables projects, CEO messaging, PA, and dashboard tools — the sweet spot for Claude Desktop users. “Read Only” is for monitoring. “Full Access” unlocks everything.

The key is scoped to your account, not to a specific AI client. This means you can use the same key in Claude Desktop, a ChatGPT custom GPT, and a Gemini function calling setup simultaneously. Usage is tracked per-key on the dashboard so you can see exactly which client is making which calls. If a key is compromised, revoke it instantly — active sessions are terminated and a new key can be generated in seconds.

What we did not build (yet)

HiveProtocol ships with deliberate limitations. Knowing what we left out is as important as knowing what we included.

No file uploads through AI assistants. You cannot attach a Figma file or a spec document through the chat interface. Files still go through the dashboard. The security and validation requirements for file handling are different enough from text-based tool calls that we did not want to rush it.

No payment initiation. You can create a project and complete intake through your AI assistant, but funding still happens through our secure checkout on the dashboard. We are not comfortable routing payment flows through third-party AI interfaces. Maybe someday, but not today.

No streaming responses. Tool responses from HiveProtocol are returned as complete payloads, not streamed. For short responses (status checks, intake questions) this is fine. For longer responses (detailed agent messages, dashboard summaries) there is a noticeable pause. We plan to add streaming support as MCP and function calling protocols mature.

Getting started

Sign up for HiveClaw (or log in to your existing account). Navigate to Settings → HiveProtocol. Generate an API key with the scopes you need. Then follow the setup guide for your AI assistant — we have step-by-step docs for Claude Desktop, ChatGPT, and Gemini.

Your best ideas happen in conversation. Now they do not have to die there.