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

# Run Vise as an MCP server

> Expose Vise's local tools to Claude Code, Cursor, Codex, and VS Code over MCP.

Vise ships an MCP adapter, so any MCP-capable host can call Vise's tools — inspect, plan, check, sensors, docs lookup, and more — directly during the coding loop. Start it with:

```sh theme={null}
vise mcp
```

This runs a local **stdio** MCP server. You normally don't run it by hand; you register the command in your host's MCP configuration and the host launches it for you.

<Info>
  This is different from the hosted **docs** MCP server at `https://learn.social.plus/mcp` ([overview](/ai-mcp/overview)). That one is remote (SSE) and only reads documentation. The Vise MCP server runs **locally** and exposes Vise's validation and planning tools against the repository you're working in. You can use both at once.
</Info>

## Prerequisites

Install Vise so the `vise` command is on your `PATH`:

```sh theme={null}
npm install -g @amityco/social-plus-vise
vise doctor
```

If you prefer not to install globally, replace the command below with `npx` and args `["-y", "-p", "@amityco/social-plus-vise", "vise", "mcp"]`.

## Configure your host

<Tabs>
  <Tab title="Claude Code">
    Add a project-scoped `.mcp.json` in your repository root:

    ```json theme={null}
    {
      "mcpServers": {
        "social-plus-vise": {
          "command": "vise",
          "args": ["mcp"]
        }
      }
    }
    ```
  </Tab>

  <Tab title="Cursor">
    Add `.cursor/mcp.json` in your project root:

    ```json theme={null}
    {
      "mcpServers": {
        "social-plus-vise": {
          "command": "vise",
          "args": ["mcp"]
        }
      }
    }
    ```
  </Tab>

  <Tab title="VS Code">
    Add `.vscode/mcp.json` in your project (VS Code uses a `servers` key):

    ```json theme={null}
    {
      "servers": {
        "social-plus-vise": {
          "command": "vise",
          "args": ["mcp"]
        }
      }
    }
    ```
  </Tab>

  <Tab title="Codex">
    Add the server to `~/.codex/config.toml`:

    ```toml theme={null}
    [mcp_servers.social-plus-vise]
    command = "vise"
    args = ["mcp"]
    ```
  </Tab>
</Tabs>

Restart or reload the host after editing its configuration, then confirm `social-plus-vise` appears as an available MCP server.

## What it exposes

The Vise MCP tools mirror the [CLI commands](/ai-mcp/vise/cli-reference), named in snake\_case. The ones you'll use most:

| Tool                           | Mirrors                                  | Purpose                                      |
| ------------------------------ | ---------------------------------------- | -------------------------------------------- |
| `inspect_project`              | `vise inspect`                           | Detect platform, surfaces, and sensors       |
| `plan_integration`             | `vise plan`                              | Grounded plan with docs citations and intake |
| `init_compliance`              | `vise init`                              | Write the `sp-vise/` contract                |
| `check_compliance`             | `vise check`                             | Validate the code against the contract       |
| `run_sensors`                  | `vise run-sensors`                       | Run build, lint, typecheck, and smoke checks |
| `get_sdk_facts`                | `vise sdk-facts`                         | Read grounded SDK surface facts              |
| `search_docs` / `get_doc_page` | `vise search-docs` / `vise get-doc-page` | Find and read social.plus docs               |
| `debug_issue`                  | `vise debug`                             | Diagnose an SDK-specific runtime failure     |
| `attest_rule` / `explain_rule` | `vise attest` / `vise explain`           | Record evidence and read rule guidance       |

## Skill or MCP?

The [skill](/ai-mcp/vise/overview#install-the-ai-skill) and the MCP server are complementary:

* The **skill** teaches the agent *when* to inspect, plan, fetch docs, validate, and attest.
* The **MCP server** exposes the *tools* the agent calls to do it.

For MCP-capable hosts, install the skill **and** register the MCP server. On hosts without MCP support, install the skill and let the agent call the `vise` CLI directly.

## Related

<CardGroup cols={2}>
  <Card title="Vise overview" icon="shield-check" href="/ai-mcp/vise/overview">
    Install Vise and run your first integration.
  </Card>

  <Card title="How Vise works" icon="gears" href="/ai-mcp/vise/how-it-works">
    The governed loop and how to read a check result.
  </Card>

  <Card title="CLI reference" icon="terminal" href="/ai-mcp/vise/cli-reference">
    The full command set behind these tools.
  </Card>

  <Card title="Docs MCP server" icon="message-bot" href="/ai-mcp/overview">
    Connect an AI tool to the hosted documentation MCP server.
  </Card>
</CardGroup>