The Caylex Navigator

The Caylex Navigator is the core of the platform. It is an MCP server that sits between your AI agents and all of your connected MCP servers, adding intelligence, security, and observability to every tool interaction. Your agent connects to a single endpoint — the Caylex Navigator — and gains access to tools across all servers in its project. The Navigator handles tool discovery, authentication, execution, and analytics automatically.

​

How it works

1. Your agent connects to the Caylex Navigator at https://proxy.caylex.ai/mcp using Streamable HTTP transport
2. The Navigator identifies the navigator instance and project from the x-api-key header, and the user from the x-user-email header
3. The Navigator exposes MCP tools that your agent uses to discover and execute tools across all connected servers
4. When your agent invokes a tool, the Navigator automatically injects the correct credentials for that user and server, executes the call, and returns the result
5. Every interaction is logged for analytics and debugging

​

The Navigator’s MCP tools

The Caylex Navigator exposes up to 5 MCP tools to your agent, depending on your Navigator Instance configuration:

​

suggest_tools

Recommends the most relevant tools based on your agent’s intent. Your agent describes what it wants to accomplish, and the Navigator returns a ranked list of tools with descriptions and relevance scores. This tool is available when Dynamic Suggestions is enabled on the Navigator Instance.

​

get_tool_schemas

Retrieves the input schemas for one or more tools. After your agent selects tools (from suggestions or direct knowledge), it calls this tool to get the parameter schemas needed to invoke them. Schemas are fetched in parallel across multiple servers for efficiency.

​

invoke_tools

Executes one or more tool calls in parallel. Your agent provides the tool name, server name, parameters, and an intent description. The Navigator:

• Looks up the correct server and tool
• Retrieves and injects the user’s credentials automatically
• Executes the tool call on the external MCP server
• Returns the result (or an enriched error message)

Multiple tool calls are executed in parallel for performance.

​

get_context_map

Returns a semantic overview of what information exists across all connected servers. This helps your agent understand where data lives before making queries. This tool is available when Context Maps is enabled on the Navigator Instance.

​

get_authentication_link

Returns an authentication link that users can use to authenticate with servers they haven’t connected to yet, directly from within a conversation. This tool is available when In-Chat Authentication is enabled on the Navigator Instance.

​

Automatic authentication

One of the Navigator’s key capabilities is automatic credential injection. When your agent invokes a tool, the Navigator:

1. Identifies which server the tool belongs to
2. Looks up the user’s authentication credentials for that server (based on the x-user-email header)
3. Injects the credentials into the request (OAuth token, API key, path/query parameters — depending on the server’s auth method)
4. Connects to the external MCP server and executes the tool

Your agent never needs to handle credentials directly. The Navigator manages the entire authentication flow transparently.

​

Error recovery

When a tool call fails, the Navigator enhances the error message with historical context. It retrieves examples of previous successful calls to the same tool, including:

• The parameters that were used
• A truncated version of the successful result

This context helps your agent understand what a correct call looks like and self-correct on the next attempt.

​

Observability

Every interaction through the Navigator is tracked:

• Sessions — each session is logged with all tool calls
• Tool call metrics — latency, success/failure, error types
• User activity — which users are using which tools

See Analytics for details on monitoring your deployments.