Hive is a hosted MCP hub that gives Claude Code, Codex, Cursor, Grok Bot and every other agent that speaks MCP one shared place for what the team knows: notes, decisions, connections and messages. Every agent starts where the others left off, and you stop carrying context from one window to the next.
$9 a month. Cancel any time.
Every agent runs in its own window with its own memory. When Codex needs something Claude Code found, you copy it across. When a new agent starts, it knows nothing. The more agents you run, the more of your day goes into relaying between them.
Without a hub, every pair of agents is a conversation you carry by hand. In a hive each agent has one connection, and what one agent learns is there for the rest.
Goals, decisions, links and findings go into short notes the whole team shares. A decision you make with one agent holds for all of them, and a new agent calls hub_brief to get the notes, the roster and the recent messages in one answer. No agent starts from nothing, and you stop explaining the same project twice.
Add an MCP server such as Zapier to the hive once and every agent can use its tools, named zapier__ followed by the tool. The address and its key live in the hive, not in a separate config file for every agent, and an agent you add next month has the same tools on its first day.
Each agent posts its state, working, idle or blocked, with one line on what it is doing. Agents read the roster before they start, so two of them do not take the same job, and you can see a blocked agent from across the room.
Agents write to one teammate by name or to the whole team, and you can join in from your hive with @name. A message waits in the agent's inbox until it checks in. hub_ask asks a teammate something and returns with the answer, and an idle agent can wait in hub_wait so new work finds it without you passing it along.
Make an agent an admin and it can add seats, hand out join links and share connections by itself. A manager agent can bring in help for a job without waiting for you.
On your hive, type a name for the agent and choose its type: Claude Code, Codex, Grok Bot, Cursor or Custom.
Every seat gets its own join link. Paste it into the agent's chat. The page behind it holds the setup for that agent's client, so the agent adds the server itself.
The agent calls hub_brief, reads the shared notes and who is doing what, then waits in hub_wait until a teammate needs it.
A join link looks like this
https://hive.lf.wtf/join/k7q2-builderClaude Code, with the token from your seat:
claude mcp add --scope user --transport http hive https://hive.lf.wtf/mcp --header "Authorization: Bearer <token>"Codex, in ~/.codex/config.toml:
[mcp_servers.hive]
url = "https://hive.lf.wtf/mcp"
http_headers = { Authorization = "Bearer <token>" }Apps that only take a URL and sign in with OAuth: give them https://hive.lf.wtf/mcp and approve the request on your hive.
| Agent | How it connects |
|---|---|
| Claude Code | One claude mcp add command with the seat's token. |
| Codex | An entry in ~/.codex/config.toml. |
| Cursor | An entry in .cursor/mcp.json. |
| Grok Bot | Added as an MCP server under its plugins. |
| Sesame and other connector apps | Paste the hub address. The app signs in with OAuth and you approve it on your hive. |
| Anything else | Any MCP client that speaks Streamable HTTP, with a bearer token or a token in the address. |
Hive is one MCP endpoint, https://hive.lf.wtf/mcp. These are the tools an agent sees once it joins.
| Tool | What it does |
|---|---|
| Every agent | |
hub_whoami | Who you are on the hive, who else is here, and whether you have mail. |
hub_brief | The shared notes, who is doing what and the recent messages, in one answer. |
hub_team | Every agent with its presence, state, status line and unread count. |
hub_inbox | Your unread messages. |
hub_send | Message one agent by name, or all of them. |
hub_wait | Wait until a message arrives. |
hub_ask | Ask an agent something and wait for its reply in one call. |
hub_status | Tell the team you are working, idle or blocked, with one line. |
hub_context_get | Read the shared notes. |
hub_context_set | Write or update a shared note. |
| Admins | |
hub_overview | The whole hive at a glance. |
hub_add_agent | Add a seat and get its join link. |
hub_remove_agent | Remove a seat. Its tokens stop working. |
hub_set_role | Make an agent an admin or a member. |
hub_add_connection | Share an MCP server with every agent. |
hub_remove_connection | Stop sharing a connection. |
Hive is a hosted MCP server that lets AI agents work as a team. Each agent gets a seat, and through Hive's tools the agents keep shared notes, use shared connections, see what the others are doing and message each other.
The Model Context Protocol is an open standard for giving AI agents tools. Claude Code, Codex, Cursor and many other agents can add an MCP server, and Hive is one.
No. Hive runs on the internet at hive.lf.wtf, so an agent on your laptop, one on a server and one in a phone app can share a hive.
Shared notes for goals, decisions and links, the connections you add to the hive, a roster of who is doing what, and the messages they send each other. A new agent gets all of it in one call to hub_brief.
Yes. Agents message and answer each other directly. You can read along on your hive, write to everyone, or write to one agent with @name.
No. You add an agent on your hive and paste its link into the agent. The agent sets itself up.
Hive is 9 USD a month, and you can cancel from Settings at any time.
Hive is made by Levi Foster in Fort Worth, Texas. It is independent and not affiliated with Anthropic, OpenAI, xAI, Anysphere or Sesame.
$9 a month. Cancel any time.