Skip to main content
MeshAgent includes a set of built-in toolkits and toolkit builders that you can attach to agents. This page is focused on the agent-facing toolkits you use from the CLI or SDK. It is intentionally an overview page, not a per-method reference. The MeshAgent CLI makes it easy to extend agents with built-in tools. For example, you can give GPT- or Claude-based agents the ability to search the web, interact with room storage, use MCP tools, create and retrieve memories, and more.
  • A toolkit is a set of tools that is attached directly to an agent.
  • A toolkit builder is something the agent can use to construct tools dynamically, often on a per-turn basis.

​
Using built-in toolkits from the CLI

Built-in toolkits are commonly enabled when you start an agent from the CLI: 
meshagent chatbot join \
  --room my-room \
  --agent-name GPTResearcher \
  --model gpt-5.4 \
  --web-search \
  --mcp \
  --require-storage \
  --use-memory agents/gptresearcher/memories

meshagent chatbot join \
  --room my-room \
  --agent-name ClaudeResearcher \
  --model claude-sonnet-4-6 \
  --web-fetch \
  --mcp \
  --require-storage \
  --use-memory agents/clauderesearcher/memories
In those cases, the CLI is attaching built-in toolkits or toolkit builders to the agent you launch. There are two common patterns:
  • Always-on tools: the tools are attached directly to the agent and are always available to the model.
  • Per-turn tools: the tools are exposed through toolkit builders, so they can be selected or toggled for a given turn in the UI.
Many built-in CLI flags support both forms. When they do, the always-on form is typically prefixed with --require-. For example:
  • --web-search makes web search available as a per-turn tool.
  • --require-web-search makes web search always available to the agent.
  • --web-fetch makes web fetch available as a per-turn tool.
  • --require-web-fetch makes web fetch always available to the agent.
  • --mcp makes MCP tools available through the normal builder flow. There is not currently a --require-mcp mode in the chatbot CLI.
  • --storage uses StorageToolkitBuilder for the builder-based flow.
  • --require-storage adds StorageToolkit directly as always-on tools.
  • --use-memory agents/<agent-name>/memories adds a MemoriesToolkit for that memory path.

​
Built-in toolkits

These are first-party MeshAgent toolkits that you attach to agents or services in code.

​
Document and content tools

  • DocumentAuthoringToolkit: create and modify MeshDocuments.
  • DocumentTypeAuthoringToolkit: work with type-specific document structures.
  • MarkItDownToolkit: convert PDFs and Office documents into formats that are easier for LLMs to use.
These are commonly used in document generation, website authoring, document intelligence, and RAG-style workflows.

​
Data and storage tools

  • StorageToolkit / StorageToolkitBuilder: expose room or mounted storage operations as tools.
  • MemoriesToolkit / MemoriesToolkitBuilder: expose room-backed memory tools such as add, search, inspect, and delete memory operations.
  • DatabaseToolkitBuilder: construct database tools for agent workflows.
  • UUIDToolkit: provide UUID generation helpers.

​
Provider and adapter-driven toolkits

These toolkits are usually exposed through adapters or builders instead of being always on. Some of them switch implementation based on the model you choose, so the same CLI flag can work across providers even when the underlying tool definition is different. For adapter-specific details, see OpenAI Responses Adapter and Anthropic Messages Adapter.
  • MCPToolkit / MCPToolkitBuilder: expose external MCP servers as MeshAgent tools.
  • WebFetchToolkitBuilder: add web-fetch tools. In the CLI, --web-fetch selects Anthropic’s built-in Messages API web fetch tool for Claude models and MeshAgent’s WebFetchToolkitBuilder for other models. That means the same flag works for both Anthropic and OpenAI-backed agents.
  • WebSearchToolkitBuilder: add web search tools. For Claude models this uses Anthropic’s built-in Messages API web_search_* tool. For other model paths the CLI selects a different builder, so the same --web-search flag still works outside Anthropic.
  • ShellToolkitBuilder: add function-based shell execution tools. In the current CLI, --shell is wired for both Claude and OpenAI model paths.
  • LocalShellToolkitBuilder: expose OpenAI’s native local shell tool. In the current CLI, --local-shell is OpenAI-only.
  • ImageGenerationToolkitBuilder: expose OpenAI-native image generation. In the current CLI, --image-generation is OpenAI-only.
  • ApplyPatchToolkitBuilder: expose OpenAI’s patch-editing tool. In the current CLI, --apply-patch is OpenAI-only.
  • ScriptToolkitBuilder: turn a fixed script or command sequence into a tool the agent can call. This is useful when you want to wrap an existing CLI workflow or shell script as an agent tool without building a full custom Python tool first.