MCP authentication and headers

Each MCP server on an agent has its own authentication, set on the Tools tab when you add or edit it. What you choose is sent on every request to that server: the handshake, tool discovery and every tool call.

Authentication types

Custom headers

Add any other headers the server expects under Custom headers, one KEY=VALUE per line, for example a workspace or version header. Headers that HTTP manages itself, such as Host, Content-Length and Transfer-Encoding, can't be set.

Headers SentientOne always sends

POST /mcp HTTP/1.1Host: mcp.example.comContent-Type: application/jsonAccept: application/json, text/event-streamAuthorization: Bearer …X-Request-Id: 284662c8227a417d9f66755385919e0e{"jsonrpc": "2.0", "id": 1, "method": "initialize", "params": {"protocolVersion": "2024-11-05", …}}# then, reusing any Mcp-Session-Id the server returned:{"jsonrpc": "2.0", "id": 2, "method": "tools/list"}{"jsonrpc": "2.0", "id": 3, "method": "tools/call", "params": {"name": "get_order", "arguments": {"order_id": "10482"}}}

• Accept: application/json, text/event-stream, so the server may answer either way.
• X-Request-Id: the correlation id of the chat request that led to the call. Log it on your server to match a tool call to its trace. See Request ids.
• Mcp-Session-Id, once your server has returned one.

How secrets are stored

• Tokens and passwords are encrypted at rest with AES-256-GCM before they are saved, and decrypted only when a request is made.
• They are never sent back to the browser in full; the Tools tab shows them masked.
• Caches hold only the encrypted value.