HelpTools

Tools glossary

Links on this page to features marked Preview describe designs that are not available yet.

Terms for tool calling, built-in tools and MCP. Shorter definitions of the core API terms are on the main Glossary.

Code execution container

The sandbox where syntic_code_execution runs code; inputs at /mnt/inputs, outputs at /mnt/outputs. See code execution.

Computer use

Tools that let the model operate a screen with mouse and keyboard actions your app carries out. See computer use.

Connector

The small process you run for a tunnel. It forwards calls only to the tunnel’s allowed_targets.

Deferred loading

defer_loading: true on a tool: it is searchable but not sent to the model until needed.

Function tool

A tool you describe with a name, description and JSON schema, and run in your own code. See tool calling.

is_error

Marks a tool result as a failure so the model can retry or explain instead of trusting it.

MCP

Model Context Protocol: an open protocol for exposing tools and data to models through a server.

MCP connector

The mcp_servers request field: Syntic connects to the listed servers and offers their tools to the model. See mcp connector.

MCP policy

The organization’s rule for which MCP servers may be used: open, allowlist or disabled. See mcp policy.

MCP tunnel

An outbound-only connection from your network that lets Syntic reach a private MCP server. See mcp tunnels.

Parallel tool calls

Several tool calls in one assistant turn. Return one result per call id. See parallel tools.

Search result block

Your own search hits passed as citable content. See search results.

Server tool

A built-in tool Syntic runs during the request, such as syntic_web_search or syntic_code_execution. See tools.

Standard tool

A published tool definition the model knows well (syntic_text_editor, syntic_shell, syntic_memory) that your app executes. See standard tools.

Strict mode

function.strict: true: arguments are guaranteed to match the schema exactly. See strict tools.

Tool result

Your reply to a tool call: a tool message with the matching tool_call_id. See tool results.

Loading tool definitions on demand from a large catalog instead of sending all of them on every call. See tool search.

tool_choice

Controls whether the model may, must, or must not call tools, or must call one named tool.

Was this page helpful?