> ## Documentation Index
> Fetch the complete documentation index at: https://docs.hoista.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP server

> How agents connect to Hoista's MCP server and which tools to start with.

Hoista runs an MCP server so agents can read and change your projects. Every tool call goes through the daemon, which checks the agent's access level and asks you when a change needs your approval. The full list of tools is in the [tool catalog](https://docs.hoista.dev/reference/mcp-tools.md).

## Connections

| Agent | How it connects | Access levels |
| - | - | - |
| Coding tool on this computer | stdio, through `hoist mcp` | **View only**, **Ask for approval**, **Full access** |
| Chat app or cloud agent | `https://<your-hoista-address>/mcp`, signed in with OAuth | **View only**, **Full access** |

A chat app or cloud agent connects over HTTP, where Hoista can't ask you in the agent's window, so it can't use **Ask for approval**. It reaches either Hoista on your computer, through a tunnel or proxy to `127.0.0.1:8788` (see [From your computer](https://docs.hoista.dev/agents/tools/chatgpt.md#from-your-computer)), or a [remote](https://docs.hoista.dev/guides/remote.md). Find the address in **Settings → Agents**.

To add Hoista to a coding tool, print its configuration:

```bash theme={null}
hoist mcp config
```

It prints an `mcpServers` entry that runs `hoist mcp` with your state folder. With a remote in use, it prints the remote's `/mcp` address instead. It contains no credentials. Add `--name NAME` to label the connection. When the tool starts a session, Hoista opens a page in your browser where you check a code and choose the access level. If the page doesn't open, the request waits in **Agents**, or answer it with `hoist agents allow`. See [Connect an agent](https://docs.hoista.dev/agents/connect.md).

## Remote endpoints

| Path | Use |
| - | - |
| `/mcp` | MCP over HTTP with an OAuth bearer token |
| `/.well-known/oauth-protected-resource` | OAuth resource metadata, found by the agent itself |
| `/.well-known/oauth-authorization-server` | OAuth server metadata |
| `/oauth/register`, `/oauth/authorize`, `/oauth/token` | Client registration, sign-in and token exchange |

## Tools to start with

| Tool | Use |
| - | - |
| `get_my_access` | What this connection may do, and which projects and environments it can reach. Call it first. |
| `list_projects` | Find the project to work on. |
| `list_provider_operations` | Everything the provider behind an account or resource can do through Hoista. |
| `describe_provider_operation` | An operation's arguments and risk. |
| `run_provider_operation` | Run one of those operations. |
| `plan_deploy`, `deploy` | Make a deployment plan, then queue it for a managed project. |
| `wait_for_deploy` | Wait until a deploy is live, failed or stopped. |
| `get_logs`, `get_log_history` | Capture live log records, or read earlier logs where the provider keeps them. |

Each client gets the tool schemas from `tools/list`. Agents can't change their own access. Only you can, from **Agents** or with `hoist agents access`.

## Related

> [card] [MCP tool catalog](https://docs.hoista.dev/reference/mcp-tools.md)
>
> Every tool an agent can call.

> [card] [Connect an agent](https://docs.hoista.dev/agents/connect.md)
>
> Every tool, and how each connects.

> [card] [Agent permissions](https://docs.hoista.dev/agents/permissions.md)
>
> Change or narrow what an agent can do.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.