Skip to main content
The Sphinx Model Context Protocol (MCP) server gives your agent tools to search your organization’s knowledge, read complete pages, and suggest changes. You can connect it to clients such as Claude and ChatGPT. Sphinx tools expose the learned organizational context stored in your workspace. For context augmentation in the inference request path, see the Responses API.

Before you start

  • A Sphinx account with access to the workspace you want to use.
  • An agent or client that supports remote MCP servers over Streamable HTTP.
  • For API-key authentication, permission to create a key in that workspace and a client that supports custom request headers.

Authenticate

Choose the method supported by your client. Both methods use the same MCP endpoint.

OAuth: sign in through your browser

1

Add the remote server

Open your client’s MCP or connector settings. Add a server using the name, transport, and URL above. Choose OAuth if the client asks for an authentication method. If it offers dynamic client registration, choose automatic registration.
2

Authenticate

Use your client’s Authenticate, Connect, or Sign in & test action. Sign in to Sphinx in the browser that opens.
3

Select your workspace

Select the Sphinx workspace you want the agent to use and review the authorization request. Click Allow access to authorize it.
4

Return to your agent

Finish saving or applying the connection in your client. Enable Sphinx for the conversation if your client requires it.

API key: send a Bearer header

1

Create a workspace key

Open Sphinx, click Settings at the bottom left, and check the selected Workspace. Open API keys, click New key, enter a name and expiration, then click Create. Copy the generated key.
2

Configure your client

Add the remote MCP endpoint and supply this request header, replacing the placeholder with your key:
For a client that uses a Claude-style .mcp.json, the configuration is:
Configuration formats vary by client. See Claude Code API key setup for the plugin-specific steps.
3

Reconnect and verify

Save the configuration and reconnect the MCP server. Confirm that the client lists Sphinx’s tools, then try a question about a page in your workspace.
Keep API keys private. Do not commit a configuration containing your key or share it in screenshots, prompts, or logs. Use your client’s secret storage when available.

Use Sphinx in a conversation

Try a question tied to your workspace’s content:
For an update, name the pages and the specific change you want. Follow the resulting job in Ingest and review the proposed changes in Sphinx.

Connect Sphinx MCP to your agent

Claude

Set up the plugin and connect Claude.ai, Claude Desktop, or Claude Code.

Claude Code with an API key

Configure Bearer authentication for the Sphinx plugin’s MCP server.

ChatGPT Desktop

Add the remote MCP server and authorize a workspace.

Claude with Private Gateway

Add MCP in Claude’s third-party inference settings.

Troubleshooting

  • Authentication fails: check the endpoint and transport. For API keys, check expiration and that the header starts with Bearer followed by the key. For OAuth, authenticate again through the client.
  • Expected knowledge is missing: check that you authorized the intended workspace or created the key there. Ask about a page you can find in Browse Knowledge.
  • Sphinx is connected but unused: enable it for the conversation and explicitly ask the agent to use Sphinx.
  • A suggested edit is not visible yet: check the ingest’s progress and review status. Submitting an update request does not mean it has been accepted.
MCP provides knowledge tools. To route model inference through Sphinx, see Sphinx Private Gateway.