Skip to main content
Caylex exposes a single MCP endpoint that your agent connects to. Through this one connection, your agent gains access to all tools across all servers in its project.

​
Endpoint and transport

SettingValue
Endpoint URLhttps://proxy.caylex.ai/mcp
TransportStreamable HTTP (MCP standard)
ProtocolModel Context Protocol (MCP)
Any MCP-compatible agent framework can connect to Caylex — no custom SDKs or adapters required.

​
Required headers

Every request to the Caylex Navigator must include two authentication headers:

​
x-api-key

Your Navigator Instance’s API key. This tells Caylex which navigator and project the request belongs to.
  • Format: ck_{id}.{secret}
  • Generated on the Navigator Instance page in the Caylex dashboard
  • See Navigators & Permissions for how to create API keys

​
x-user-email

The email address of the end user currently interacting with your agent. This tells Caylex which user’s credentials to use when executing tools.
  • Must match the email the user used when authenticating via an Auth Link
  • Determines which credentials the Navigator injects when calling external MCP servers
  • Used for per-user analytics and access control
The x-user-email must exactly match the email the user authenticated with. If it does not match, tool calls to servers requiring authentication will fail.

​
How it works at runtime

When your agent connects and makes tool calls, here is what happens:
1

Agent connects to the Navigator

Your agent opens an MCP connection to https://proxy.caylex.ai/mcp with the x-api-key and x-user-email headers.
2

Navigator identifies context

The Navigator validates the API key and resolves:
  • Which Navigator Instance the key belongs to
  • Which Project the navigator is deployed in
  • Which user the email corresponds to
  • Which tool restrictions and Navigator capabilities are configured
3

Agent discovers tools

The Navigator returns the available MCP tools based on the navigator’s configuration:
  • suggest_tools (if Dynamic Suggestions is enabled)
  • get_tool_schemas
  • invoke_tools
  • get_context_map (if Context Maps is enabled)
  • get_authentication_link (if In-Chat Authentication is enabled)
4

Agent invokes tools

When the agent calls invoke_tools, the Navigator:
  1. Looks up the target server and tool
  2. Retrieves the user’s credentials for that server
  3. Injects the credentials into the request
  4. Executes the tool call on the external MCP server
  5. Returns the result to the agent

​
Integration examples

Choose your agent framework for a complete integration guide:

OpenAI Agents SDK

Python integration using the OpenAI Agents SDK.

LangChain

Python integration using LangChain MCP Adapters.

Claude Agent SDK

Python integration using the Claude Agent SDK.

Claude Desktop

JSON configuration for the Claude Desktop app.

​
Environment variables

We recommend storing your credentials as environment variables rather than hardcoding them: 
export CAYLEX_API_KEY="ck_abc123.your-secret-key"
export CAYLEX_USER_EMAIL="user@example.com"
Never commit API keys to version control. Use environment variables, secret managers, or .env files (with .gitignore) to manage credentials.

​
Any MCP-compatible framework

Caylex works with any framework that supports the MCP Streamable HTTP transport. The connection pattern is always the same:
  1. Point at https://proxy.caylex.ai/mcp
  2. Set the x-api-key and x-user-email headers
  3. Let the framework handle MCP tool discovery and invocation
If your framework supports MCP but is not listed in our integration guides, follow its documentation for connecting to a remote MCP server with custom headers.