Skip to main content
Hive is a Model Context Protocol (MCP) client. If a tool publishes an MCP server, Hive can connect to it, discover its tools, and use them in Ask, workflows and agents — with the same approvals and audit trail as every other connector.
Hive connects to MCP servers. It does not publish its own MCP server for other AI tools to call.

​
Official servers: one-click connect

These vendor-hosted servers are verified to work with Hive and connect with OAuth in one click: Find them on the provider’s tile in Integrations → Connectors. Tools that offer both OAuth and MCP (Notion, Linear, Stripe) show one tile with a choice of connect method.

​
Connect a custom server

1

Open Add a connector

In Integrations → Connectors, select Add integration, then Custom MCP server — “Any Model Context Protocol server by URL.”
2

Enter the server details

Give the server a name and its URL. The URL must be https://. Hive speaks Streamable HTTP only; it never runs local stdio servers.
3

Choose how it authenticates

  • No auth — a public server with no sign-in.
  • Bearer token — paste a static token. Hive seals it in its vault; it is never stored on the connection record or shown again.
  • OAuth 2.1 — OAuth 2.1 with PKCE and dynamic client registration. Optionally list the scopes to request, space or comma separated. You are sent to the server’s sign-in page and returned to Hive.
4

Let Hive discover the tools

Once connected (status Live), Hive lists the server’s tools and pins each one. The server appears under the Built by me filter.

​
How Hive keeps MCP tools safe

MCP servers are third-party code, so Hive treats every tool as untrusted until you decide otherwise:
  • Tools are pinned. Hive records a fingerprint of each tool’s name, description and input schema when it discovers it. If a server later changes a tool behind your back — the “rug-pull” attack — the change is caught rather than silently trusted. Malformed tools are skipped with a reason.
  • Tools are namespaced per connection, so two servers can never collide or impersonate each other.
  • High-risk by default. Every MCP tool starts as blast radius B3 (external, hard to reverse), not reversible and not read-only. Hive ignores a server’s own claim that a tool is read-only; only an operator override can lower the tier.
  • Reads vs writes. On the audited official servers, Ask may call read tools automatically while answering your question. On custom servers, reads — and on every server, all writes — go through approval.
  • Secrets are sealed. Bearer and OAuth tokens are encrypted in Hive’s vault and never given to the model.
See Autonomy and blast radius for what B3 means.

​
Statuses

​
Resync and disconnect

  • Re-sync re-runs discovery and re-pins the server’s tools — use it after the server adds or changes tools you want to use. Only the connection’s owner can resync.
  • Disconnect drops the connection, its pinned tools and all of its sealed secrets.
Owners and Admins can see metadata for every MCP server connected in the workspace, but not the credentials.

Incoming webhooks

Push events from any system.

Security overview

How Hive keeps actions safe.