> ## Documentation Index
> Fetch the complete documentation index at: https://speaker-weave.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Hosted MCP server

> Connect ChatGPT, Claude, or any MCP client to your SpeakerWeave organization.

Every deployment hosts a remote MCP server at `/mcp` (Streamable HTTP) exposing 16 organization-scoped tools and 4 resources: events, per-event branding, submissions, decisions, speakers, portal invitations, scheduling, content status, reminders, evaluation summaries, and AI triage.

There is one thing to paste into any client:

```
https://speakerweave.com/mcp
```

Self-hosting? Use your own origin with the same `/mcp` path.

## Set it up in ChatGPT

Custom MCP connectors are available on ChatGPT Plus, Pro, Business, Enterprise, and Edu; on Business and above a workspace owner may need to allow custom connectors first.

<Steps>
  <Step title="Open connector settings">
    **Settings → Connectors** (some plans label this **Apps & connectors**), then **Create** / **Add custom connector**.
  </Step>

  <Step title="Point it at SpeakerWeave">
    Name it `SpeakerWeave`, paste `https://speakerweave.com/mcp` as the MCP server URL, and choose **OAuth** as the authentication method. No headers, no client ID, no secret — the server publishes its own metadata and registers the client for you.
  </Step>

  <Step title="Authorize">
    ChatGPT opens SpeakerWeave's approval page. Sign in, pick the organization, and approve. You will need an API token from **Settings → API tokens** in SpeakerWeave; the approval page exchanges it for short-lived OAuth access, so the raw token never reaches the client.
  </Step>

  <Step title="Use it in a chat">
    Start a new chat, enable **SpeakerWeave** from the tools/connectors picker, and ask something real:

    ```
    What's still pending review for the AI Builders Summit?
    ```

    ChatGPT calls `list_submissions` and answers from your live program.
  </Step>
</Steps>

## Set it up in Claude

Works in claude.ai, Claude Desktop, and Claude for Work. On a Team or Enterprise plan, a primary owner adds the connector for the whole organization from the same screen.

<Steps>
  <Step title="Open connector settings">
    **Settings → Connectors → Add custom connector**.
  </Step>

  <Step title="Point it at SpeakerWeave">
    Name it `SpeakerWeave` and paste `https://speakerweave.com/mcp`. Leave the advanced OAuth fields empty — discovery fills them in.
  </Step>

  <Step title="Authorize">
    Click **Connect**. SpeakerWeave's approval page opens; approve with an API token from **Settings → API tokens**. Claude stores a rotating refresh token, not your API token.
  </Step>

  <Step title="Use it in a chat">
    Open the tools menu in a new chat, switch **SpeakerWeave** on, and ask:

    ```
    Who still owes content before Aug 27?
    ```
  </Step>
</Steps>

### Claude Code

Claude Code adds the same server from the terminal:

```bash theme={null}
claude mcp add --transport http speakerweave https://speakerweave.com/mcp
claude mcp list        # confirms the connection
```

Then `/mcp` inside a session runs the OAuth flow. The [`sw` CLI](/ai/cli) is the other way to work from a terminal — it wraps the same tool layer as a command, which is what coding agents usually reach for.

## Header-capable clients (bearer)

For any client that supports custom headers, skip OAuth entirely:

```json theme={null}
{
  "mcpServers": {
    "speakerweave": {
      "type": "http",
      "url": "https://speakerweave.com/mcp",
      "headers": { "Authorization": "Bearer dais_your_api_token" }
    }
  }
}
```

## How the OAuth flow works

The client discovers `/.well-known/oauth-protected-resource`, registers dynamically, and runs authorization-code + PKCE. SpeakerWeave's approval page exchanges an API token (from **Settings → API tokens**) for short-lived OAuth access with rotating refresh tokens — the raw API token is never stored by the client.

Tool calls are scoped to the organization that owns the token; cross-organization access is structurally impossible.

## Event branding

* `get_event_branding(event)` returns the complete resolved branding document plus the event ID, slug, and public URL.
* `set_event_branding(event, …)` merge-patches the supplied colors, fonts, layout, density, header, radius, and powered-by fields. Its generated schema advertises every allowed enum value; logo and favicon assets remain upload-only through the organizer API.
* `dais://events/{event}/branding` exposes the same resolved document as the `event-branding` JSON resource.

Both tools resolve the event inside the authenticated token's organization. An event belonging to another organization is indistinguishable from a missing event.

<Note>
  Connectors run the other way too: bring your own MCP servers *into* the SpeakerWeave agent under **Settings → Integrations → MCP connectors** ([details](/integrations/mcp-connectors)).
</Note>

## Troubleshooting

| Symptom                                  | Fix                                                                                                                             |
| ---------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- |
| Client rejects the URL                   | Include the scheme and the `/mcp` path, with no trailing slash                                                                  |
| Authorization page never loads           | The deployment must be reachable over HTTPS on a public origin — `localhost` works only for clients running on the same machine |
| Connected, but no tools listed           | Re-run the connection: the token behind it was revoked or belongs to a deleted organization                                     |
| Tools return another organization's data | Not possible — every call is scoped to the token's organization. Check which organization you approved                          |