Research tools
Discover research tools and expose them through your product's existing agent runtime.
Discover the tool catalog#
Use MCP tools/list with pagination to discover tool names, descriptions, input schemas, and annotations.
Adapt those definitions to the product's existing agent tool interface.
Keep model-provider credentials and an LLM subscription outside the Dativue connection itself.
Refresh discovery after reconnecting or a tool-list change; do not hardcode the current tool count as a permanent contract.
Execute in trusted customer context#
The backend executor selects the connection from trusted customer and conversation context, checks the product's permissions, validates arguments, and uses MCP tools/call.
Expose only the tool name and its input arguments to the model, never a token, customer selector, HTTP target, or arbitrary header field.
Apply the same customer isolation and authorization to scheduled agent work.
Preserve research results#
Return the original MCP envelope or a lossless equivalent, preserving content, structuredContent, isError, source URLs, timestamps, and result status.
Treat returned public content as untrusted research data, not instructions.