Skip to main content

How to Connect the Spyne MCP to Claude, ChatGPT, Gemini, Dust, and Microsoft Copilot

This guide explains how to manually add the Spyne MCP server to Claude, ChatGPT, Gemini, Dust, and Microsoft Copilot if it is not yet listed in their predefined connectors. MCP URL: https://mcp.spyne.app/mcp

Claude (Claude.ai and Claude Desktop)

Claude natively supports adding a remote MCP via the Custom Connectors feature. Available on Free (1 connector), Pro, Max, Team, and Enterprise plans.

Prerequisites

  • An active Claude plan (Free is sufficient for 1 connector).

  • Team/Enterprise plans: only an Owner can add a custom connector at the organization level. Members cannot add custom connectors on their own.

Steps (individual accounts: Free, Pro, Max)

  1. Open Claude.ai → click your profile → Settings.

  2. Go to Connectors in the left menu.

  3. Click the Customize link on the page.

  4. Click the + icon in the Connectors menu and select Add a custom connector.

  5. Fill in:

  6. Click Add.

  7. Complete the OAuth authentication flow if prompted.

Steps (Team / Enterprise plans)

  1. An Owner goes to Organization Settings → Connectors → Add → Custom.

  2. Enter the name and URL https://mcp.spyne.app/mcp, then publish the connector.

  3. Each member then goes to Customize → Connectors, finds "MCP Spyne", clicks Connect, and completes the OAuth flow.

Enabling in a conversation

  • In a chat, click the "+" button (bottom left) → Connectors → toggle on MCP Spyne.

  • The connector is enabled on a per-conversation basis.

Important notes

  • The MCP server must be publicly accessible on the internet. Claude connects from Anthropic's cloud infrastructure, not from your local machine. Connection is not possible behind a VPN or private firewall unless Anthropic's IPs are whitelisted.

  • To modify the configuration, you must delete the connector and re-add it.

ChatGPT

ChatGPT supports adding a remote MCP server via Custom Connectors by enabling Developer Mode.

Prerequisites

  • A Plus, Pro, Business, Enterprise, or Edu account.

  • Important limitation (since Dec. 2025): Plus and Pro plans are limited to read-only connectors (fetch/read-only). Write actions are reserved for Business / Enterprise / Edu workspaces.

Step 1 — Enable Developer Mode

  1. Open ChatGPT → Settings.

  2. Go to Apps → Advanced settings.

  3. Enable the Developer Mode toggle and confirm the warning.

Business / Enterprise / Edu: an admin must publish the connector from Workspace Settings → Connectors → Create. Members then connect individually.

Step 2 — Add the Spyne connector

  1. Go to Settings → Apps → Create app.

  2. Fill in:

    • Name: Spyne

    • Description (optional): e.g. "MCP Spyne — access to Spyne data and integrations"

    • MCP Server URL: https://mcp.spyne.app/mcp

    • Authentication: OAuth

  3. Check "I understand and want to continue" then click Create.

  4. Complete the OAuth flow.

Step 3 — Use the connector in a conversation

  1. Start a new chat.

  2. Click "+" in the input area → More → Developer mode.

  3. Select Spyne as the source for the conversation.

  4. For each tool call, ChatGPT will display the proposed call — click Confirm to authorize.

Important notes

  • Developer Mode is in beta. It must be enabled by an admin for workspaces.

  • Write actions require manual confirmation at each call (unless you memorize the approval for the conversation).

Gemini

⚠️ Important: The Gemini consumer web app (gemini.google.com) does not natively support adding a custom MCP. Remote MCPs connect via Gemini CLI or Gemini Enterprise (Google Cloud).

Option A — Gemini CLI (recommended for individual use)

Prerequisites:

Gemini CLI installed.

Steps:

1. Open (or create) the file ~/.gemini/settings.json.

2. Add the mcpServers block with a single URL (httpUrl). No need to separate the OAuth URL and the Streamable HTTP URL — Gemini CLI automatically derives the OAuth metadata endpoint (RFC 9728 §3.1) from this URL:

json

{
  "mcpServers": {
    "spyne": {
      "httpUrl": "https://mcp.spyne.app/mcp"
    }
  }
}

How does OAuth discovery work? Gemini CLI queries https://mcp.spyne.app/.well-known/oauth-protected-resource/mcp (path-suffixed form per RFC 9728). The Spyne server exposes this route as well as the root form /.well-known/oauth-protected-resource for clients that don't implement the path-suffixed form. There is no settings field to manually point to a different discovery endpoint — only httpUrl is used.

To customize scopes or timeout:

json

{
  "mcpServers": {
    "spyne": {
      "httpUrl": "https://mcp.spyne.app/mcp",
      "oauth": {
        "scopes": ["read", "write"]
      },
      "timeout": 30000
    }
  }
}

3. Restart Gemini CLI and type /mcp in the chat to verify that the Spyne MCP appears in settings (status: disconnected).

4. Trigger the OAuth connection from the chat:

/mcp auth spyne

5. Type /mcp again to confirm that Spyne tools are discovered (status: connected).

Best practices:

  • Avoid underscores in the server name (spyne, not spyne_mcp) — Gemini CLI uses them as internal separators.

  • Do not add a trailing slash to httpUrl (/mcp, not /mcp/) — some RFC 9728 probes reconstruct the URL and a double slash can break discovery.

Option B — Gemini Enterprise (via Google Cloud)

To integrate Spyne into Gemini Enterprise (Standard, Plus, or Frontline edition):

  1. In the Google Cloud Console, go to the Gemini Enterprise page.

  2. Go to Data stores → Create data store.

  3. Search for Custom MCP Server in the sources list and click Add MCP server.

  4. Fill in:

    • URL: https://mcp.spyne.app/mcp

    • Client ID / Client Secret: obtain from your IdP (Okta, Azure AD, Google, etc.) after registering Gemini Enterprise as an OAuth application.

    • MCP Server Description: describe what Spyne is used for, to help Gemini decide when to use it.

  5. Complete the configuration (region, connector name) then click Create.

  6. Wait for the data store status to switch to Active.

  7. Enable tools: Actions → Reload custom actions, then manually enable the desired tools (disabled by default).

Limitations:

  • Only Streamable HTTP transport is supported (no legacy SSE).

  • No Private Service Connect support at this time.

  • Requires the Discovery Engine Editor IAM role to create the data store.

Option C — Gemini Web App

No native support today. Multiple threads in the Gemini Apps Community confirm that Gemini CLI or Gemini Enterprise is required. Monitor Google announcements for a future web-based option.

Dust

Dust natively supports adding a remote MCP server at the workspace level. Configuration discovery and the OAuth flow are automatic from the server URL alone.

Prerequisites

  • Be a Dust workspace administrator (adding an MCP tool is done in Spaces → Tools, reserved for admins).

  • Or be on a personal workspace.

Steps

  1. Open Dust → go to Spaces → Tools.

  2. Click Add Tool → choose Add MCP Server.

  3. Enter: MCP Server URL: https://mcp.spyne.app/mcp

  4. Choose the Automatic (Recommended) authentication mode: Dust automatically discovers the OAuth configuration (RFC 9728) from the URL and initiates the connection flow.

  5. A "Configure MCP Server — How do you want to connect?" window appears. Choose a connection mode:

    • Personal accounts (recommended): each member connects with their own Spyne credentials when using the tool. This is the appropriate mode for the Spyne MCP — it preserves one OAuth session per user (and therefore each user's own rights/workspaces).

    • Shared account: all members use the credentials you provide here (a single shared Spyne account).

    • Leave the "Use custom headers" toggle disabled: OAuth auto-discovery does not require any custom headers.

    • Click Setup connection.

  6. Complete the OAuth flow when prompted.

  7. After a short sync, the "Spyne" tool is created and its tools are available.

Using the MCP in an agent

Add the "Spyne" tool to an agent's tools (Agent → Tools / Actions), then query the agent in natural language (e.g. "List the active integrations on my Spyne workspace").

Microsoft Copilot (Microsoft 365 Copilot & Copilot Studio)

Microsoft offers two ways to connect a remote MCP server to the Copilot ecosystem, depending on the intended use: conversational access in Microsoft 365 Copilot Chat (like Claude, ChatGPT, or Gemini), or integrating Spyne's tools into a custom agent built with Copilot Studio.

Option A — Microsoft 365 Copilot (custom federated connector)

Lets users query the Spyne MCP directly from Microsoft 365 Copilot Chat, in natural language.

Prerequisites:

  • Global Administrator or AI Administrator role in the Microsoft 365 admin center.

  • Access to the Teams Developer Portal.

  • Publisher information: developer name, documentation URL, privacy policy, terms of use.

Steps:

  1. Configure authentication beforehand, in the Teams Developer Portal:

    • Microsoft Entra SSO (recommended if Spyne users are already in the Microsoft tenant): register an SSO client, which generates a Registration ID.

    • OAuth 2.0 (third-party identity provider): Tools → OAuth Client Registration → + New OAuth connection, fill in Client ID, Client secret, Authorization endpoint, Token endpoint, Refresh endpoint, and Scopes, plus the redirect URI https://teams.microsoft.com/api/platform/v1.0/oAuthRedirect. Copy the generated Registration ID.

  2. Sign in to the Microsoft 365 admin center → Copilot → Connectors.

  3. Gallery tab → "Created by your org" section → Create a new connector → Add.

  4. On the Custom connector page, under "Connect to MCP server", click Add.

  5. Fill in:

  6. Click Save.

  7. (Recommended) Do a Staged rollout to a pilot group before selecting Deploy to all users.

Important notes:

  • A federated connector only exposes read-only tools (no write actions).

  • Configuration changes can take up to 15 minutes to propagate.

  • The MCP server must be accessible over HTTPS.

Option B — Microsoft Copilot Studio (custom agent)

To integrate Spyne's tools into a dedicated Copilot Studio agent rather than the general Copilot Chat.

Prerequisites:

  • Access to Copilot Studio with a license covering the standard harness.

Steps:

  1. In the relevant agent, go to Tools → Add a tool → New tool → Model Context Protocol.

  2. Fill in:

    • Server name: Spyne

    • Server description: free-text description of the use case (helps the orchestrator decide when to call the tool).

    • Server URL: https://mcp.spyne.app/mcp

  3. Choose the authentication method:

    • OAuth 2.0 → Dynamic discovery (simplest: automatic endpoint discovery via RFC 9728), or

    • Manual if automatic discovery fails: fill in Client ID, Client secret, Authorization URL, Token URL, Refresh URL, and Scopes, then register the generated callback URL with your IdP.

  4. Click Create, then Create a new connection (or reuse an existing connection) → Add to agent.

Important notes:

  • Only Streamable HTTP transport is supported (legacy SSE has been deprecated since August 2025).

  • Access is subject to Power Platform Data Loss Prevention (DLP) policies.

Summary

Plateform

Native custom MCP support

Where to configure

Authentication

Claude.ai / Desktop

✅ Yes (all plans)

Settings → Connectors → Add custom connector

OAuth (recommanded)

ChatGPT

✅ Yes (Plus+, beta)

Settings → Apps & Connectors (Developer Mode requis)

OAuth only

Gemini CLI

✅ Yes

~/.gemini/settings.json

OAuth auto-discovery

Gemini Enterprise

✅ Yes

Google Cloud Console → Gemini Enterprise → Data stores

OAuth via IdP (client_id/secret)

Gemini Web (gemini.google.com)

❌ No

—

—

Dust

✅ Yes

Spaces → Tools → Add MCP Server

OAuth auto-discovery

Microsoft 365 Copilot

✅ Yes (federated connector)

Microsoft 365 admin center → Copilot → Connectors

Entra SSO / OAuth 2.0

Copilot Studio (agent)

✅ Yes

Agent → Tools → Add a tool → MCP

OAuth 2.0 (dynamic discovery) / API key / none

Troubleshooting checklist

Before debugging, verify the following:

  1. Is the URL correct? Test over HTTPS: curl -I https://mcp.spyne.app/mcp

  2. Is the server publicly accessible? No VPN or firewall blocking Anthropic / OpenAI / Google IPs.

  3. HTTPS is required on all platforms (HTTP is not accepted).

  4. OAuth is correctly configured on the Spyne server side: /.well-known/oauth-protected-resource and /.well-known/oauth-authorization-server endpoints are accessible, and https://claude.ai/api/mcp/auth_callback is in the allowed redirect URIs if applicable.

  5. Test with a simple prompt once connected (e.g. "List the active integrations on my Spyne workspace") to confirm that tools are correctly discovered.

Did this answer your question?