Skip to main content
MCP is a standard way for AI coding assistants and chat clients to securely read data from an app like CallVault — in plain terms, it lets a tool like Claude Code or Cursor pull your call transcripts and summaries into its own conversation, instead of you copying and pasting them by hand. CallVault’s AI Connectors tab lets you connect any MCP-compatible client to your organization or a single workspace.

​
Connecting an AI client with OAuth

OAuth is the recommended way to connect — it doesn’t involve copying a token around, and the client can be revoked at any time.
1

Open AI Connectors settings

Go to Settings → AI Connectors.
2

Connect your client

Follow your AI client’s instructions to connect to CallVault via MCP over HTTP, pointing it at your CallVault organization or workspace endpoint.
3

Authorize

Approve the connection when your AI client opens the authorization flow. The connected client appears in the AI Connectors list with its name, scope, and last-used time.
OAuth connections use the endpoint URL only — do not add an Authorization header. If your client reports an error about a stale Authorization header for an OAuth connector, remove the connector entry and reconnect without one.

​
Connecting with a manual token

Some clients don’t support CallVault’s OAuth flow yet. For those, create a manual bearer token instead:
1

Create a scoped token

In Settings → AI Connectors, click Create scoped token. Choose the client you’re setting up (Claude Code, Cursor, VS Code, Windsurf, or Generic MCP), give the token a name, and choose an organization and scope (organization-wide or a single workspace).
2

Copy the setup

After creating the token, CallVault shows the token value, the MCP endpoint URL, and a ready-to-paste setup snippet for the client you selected.
3

Paste it into your client

Follow the client-specific instructions below.

​
Claude Code

Run the command CallVault generates, which looks like:
Then verify with claude mcp list.

​
Cursor

Add the generated JSON snippet to .cursor/mcp.json (project-level) or ~/.cursor/mcp.json (global), preserving the Authorization header exactly as generated.

​
VS Code

Add the generated JSON snippet to .vscode/mcp.json or your VS Code MCP settings.

​
Windsurf

Add the generated JSON snippet to ~/.codeium/windsurf/mcp_config.json.

​
Generic MCP client

Use the generated JSON snippet — it follows the standard mcpServers shape most MCP clients expect, with the endpoint URL and a bearer Authorization header.

​
Controlling what a token can access

Each manual token can be scoped to specific tool categories: Toggle categories off for a token to restrict what that client can do — useful for giving an AI assistant read-only access to your calls without write or admin permissions.

​
Managing connections

From Settings → AI Connectors you can:
  • Revoke an OAuth-connected client — CallVault immediately rejects future requests from it
  • Regenerate a manual token — the old token stops working immediately and a new one is issued
  • Delete a manual token entirely
If Claude Code shows a connection as broken after reconnecting, run claude mcp remove "<name>" -s claudeai to clear the stale entry before reconnecting.