MCP

MCP (Model Context Protocol) is a common way to give an AI agent tools: search a documentation site, query a database, file an issue, call a service you use. Each tool set is offered by an MCP server. In Avibe you connect a server once. After that, every Agent can call its tools: Vibey, Claude Code, Codex, and OpenCode alike.

What Avibe does with a server

One list for all Agents. Avibe keeps the servers in one place and tells each Agent which ones exist. An Agent looks up a server’s tools when a task calls for them and runs them through the vibe mcp command.
Servers you already have are picked up. If you configured MCP servers for Claude Code, Codex, or OpenCode on this computer, Avibe finds them and lists them too. Listing and adding never change those tools’ own files. Deleting such a server in Avibe does remove it from that tool’s file, and the confirmation says so.
Global or per project. A server can be available everywhere, or only in one project.

Add a server

Open Skills in the Workbench, choose the MCP tab, and select Add MCP. There are two ways:

Import JSON. Paste the mcpServers block a server’s own documentation gives you.
Fill in. Enter a Name and a URL or command. A web address makes an HTTP server; anything else is a command that Avibe starts on your computer. Add Headers or Environment variables if the server needs them.

As the dialog says: Once added, every Agent can call it. The server’s card then lists its tools. Avibe supports servers that run as a local command (stdio) and servers reached over Streamable HTTP. Servers that only speak the older SSE transport are not supported. Only the Owner of the Avibe can add, configure, approve, or delete servers.

When a server needs something from you

A server’s card shows what is missing:

The card says	What it means	What to do
Needs login	The server asks you to sign in with its provider. Its tools stay hidden from Agents until you do.	Select Log in and authorize in the tab that opens.
Needs config	The server’s definition names variables, such as a token, that have no value yet.	Select Configure and fill them in.
Pending approval	The server is defined in a project’s repository. Its tools stay off until an Owner approves it.	Read what it runs, then select Approve.

Servers that belong to a project

A project can carry its own servers in a .mcp.json file at the project root, in the same format Claude Code uses. Agents working in that project see them in addition to the global ones; Agents in other projects do not. Because such a file arrives with the repository, every server in it starts as Pending approval. Changing its entry in the file asks for approval again.

Where the values you enter are kept

Tokens you type into Headers, Environment variables, or Configure, and the sign-in a server gives you after Log in, are stored in Avibe’s own private files on your computer. They are not in your Vault, and Agents run as your user on the same computer. In a project’s .mcp.json, write a placeholder such as ${TOKEN} and supply the value through Configure, so that the token itself is not committed with the repository.

For Agents and the command line

Agents use, and you can run:

vibe mcp --list                 # the servers available here
vibe mcp docs --list            # the tools of the server named "docs"
vibe mcp docs search --help     # one tool's arguments
vibe mcp docs search --query "async Python"

vibe mcp-add adds a server from the command line. The full reference, including OAuth options and project files, is MCP Hub in the Avibe repository. → Related: Skills teach Agents how to do your work; MCP servers give them tools to do it with.

Start here

avibe.bot Cloud

Self-hosting

Platforms

Agents

Concepts

Reference

Compare