Smartsheet MCP server toolsets

The Smartsheet MCP server groups related tools into toolsets — logical collections of tools that share a common resource domain. Toolsets enable progressive tool loading, so that your AI client only discovers and loads the tools it needs for a given task.

AI language models have a limited context window — the maximum amount of text they can process at once. Every tool definition loaded into an AI client consumes tokens from that window.

Without toolsets, the MCP server exposes every tool at once. As the number of tools grows, this creates problems:

• Context pressure — Tool definitions compete with your conversation for space in the context window, leaving less room for the AI to reason about your data.
• Decision fatigue — The more tools available at once, the harder it is for the model to choose the right one, increasing the chance of errors or irrelevant tool calls.
• Slower responses — Processing a large tool list adds latency to every interaction, even when only a handful of tools are relevant.

Toolsets solve these problems by letting the AI client load tools on demand, keeping the active tool set small and focused.

Toolset-based tool loading follows a two-step search-then-execute pattern:

When the AI client determines it needs tools from a particular domain, it calls the search_tools tool with:

• toolset — The name of the toolset to search.
• access_type (optional) — Filter by read or write to narrow results further.

The search_tools tool returns a list of tool definitions within that toolset, including each tool's name, description, parameters, and which executor to use.

Instead of calling tools by name directly, the AI client calls the toolset's executor — a single entry point for all tools in the group. Each toolset has up to two executors:

• Read executor (e.g., execute_{toolset}_read) — For read-only operations that retrieve data without modifying it.
• Write executor (e.g., execute_{toolset}_write) — For operations that create, update, or delete data.

The executor takes:

• tool_name — Which tool to run within the toolset.
• arguments — The tool's parameters, passed as a JSON object.

The executor validates the tool name, routes the call to the correct tool, and returns the result — all through a single tool invocation.

• Reduced context usage — Only the tools relevant to the current task are loaded into the context window. A conversation about charts doesn't need to carry sheet-editing or portfolio tools.
• Better tool selection — With fewer tools visible at once, the model makes more accurate choices and produces fewer erroneous tool calls.
• Faster interactions — Smaller tool lists mean less processing overhead per turn.

• More room for conversation — With less context consumed by tool definitions, you can have longer, more detailed conversations before hitting model limits.
• More reliable results — Focused tool visibility means the AI is less likely to call the wrong tool or misinterpret parameters.
• Consistent experience — As new tools are added to the server, they're organized into toolsets automatically, so the active context doesn't grow unboundedly.

• MCP server tools — Detailed reference for each tool, tagged with its toolset
• Load server-side skills — How to load skills that orchestrate multiple tools