Toolset
A named, curated group of tools, drawn from several MCP servers and workflows, served at one MCP endpoint so each client sees only the tools it needs.
"Toolset" is not a term from the MCP specification. It is a gateway idea: instead of connecting a client to every server, or publishing one endpoint with every tool, you publish several small endpoints, each cut to one purpose. To the client, a toolset is just an MCP server with a short tools/list.
Why that beats one big server:
- Less to misuse. A support agent's endpoint holds the support tools, not the database admin tools the same upstream server also offers. That is least privilegeLeast privilege: The security principle that each user, program or agent gets only the access it needs for its task, and nothing more. applied to the tool list.
- Fewer wrong picks and fewer tokens. An agent usually sends its tool definitions to the model with every request, where they use part of the context windowContext window: The most text, counted in tokens, that a model can take into account in one call: system prompt, conversation, tool definitions and tool results, plus the answer it writes.. A short list is cheaper, and easier for the model to choose from, than a long one. See too many tools.
- One credential per purpose. A client's token can be limited to the toolsets it may call, and revoking it affects nothing else.
Example
A team defines two toolsets on one gateway:
| Toolset | Members | Used by |
|---|---|---|
support | the tracker server's read_* tools, a "refund request" workflow | the support chat agent |
eng | the Git server's tools, the docs search server | IDE agents |
The IDE agents never see the refund workflow, and the support agent never sees delete_branch.
Common confusion
A toolset is not a security rule. It decides what a client is shown, not what each call may do. Rules on arguments, approvals and roles still belong in a tool policy that checks every call.
In L2M
In the MCP gateway, a toolset is served at /mcp/<slug>. Members are workflows with an Expose as MCP Tool trigger and registered servers, each narrowed with an optional tool glob such as read_*; server tools are named <server-slug>__<tool>. Each client token lists the toolsets it may call, and a toolset can name the project whose tool policyMCP tool policy: A set of rules, enforced outside the model, that decides whether a given agent or client may call a given MCP tool with given arguments: allow it, deny it, or hold it for a person to approve. applies.
Further reading
- MCP specification 2025-11-25: tools (modelcontextprotocol.io)
- OWASP MCP Security Cheat Sheet (owasp.org)