Sixteen tools spanning files, nodes, images, comments, components and styles is a real Figma integration, not a token gesture — it is the reason to pick this over assembling three smaller servers. Two things to configure deliberately: `execute_command` runs shell instructions on your machine, and the audio and memory tools are silently absent until you install the extras.
A collection of MCP tools grouped by capability rather than by vendor, installed as one Python package. Command execution and file operations sit alongside a full Figma API surface, MarkItDown conversion, web fetching and search, Flux image generation, audio inspection and transcription, and a small memory store with semantic recall. Audio and memory tools ship as optional extras rather than in the default install.
- A command line instruction executed on the machine — `execute_command`
- Files read, written, edited by regular expression, and directories listed in detail — `read_file_content`, `write_file_content`, `replace_in_file`, `list_directory`
- The Figma API covered properly: files and nodes, images and image fills, comments read, posted and deleted, team projects and files, components and component sets, and styles — `figma_get_file`, `figma_get_file_nodes`, `figma_get_image`, `figma_get_image_fills`, `figma_get_comments`, `figma_post_comment`, `figma_delete_comment`, `figma_get_team_projects`, `figma_get_project_files`, `figma_get_team_components`, `figma_get_file_components`, `figma_get_component`, `figma_get_team_component_sets`, `figma_get_team_styles`, `figma_get_file_styles`, `figma_get_style`
- Anything converted to Markdown through MarkItDown, from a file or a URL — `convert_file_to_markdown`, `convert_url_to_markdown`
- Web pages fetched and saved, and search through Tavily or DuckDuckGo — `get_html`, `save_html`, `search_with_tavily`, `search_with_duckduckgo`
- Images generated through the Flux API and written to a file — `flux_generate_image`
- Audio length and transcribed text from a time range within a file — `get_audio_length`, `get_audio_text`
- A memory store with a thinking log, session identity, and semantic recall — `think`, `get_session_id`, `remember`, `recall`, `forget`
- A podcast episode downloaded from XiaoyuZhouFM with optional m4a to mp3 conversion — `xiaoyuzhoufm_download`
Keys only for the services you use: `FIGMA_API_KEY`, `TAVILY_API_KEY`, `DUCKDUCKGO_API_KEY` and `BFL_API_KEY` for Flux image generation. The recommended launch is `uvx "mcp-toolbox@latest" stdio`; audio and memory tools are **not** in the default install and need the extras — `[audio]`, `[memory]` or `[all]`, as in `uvx "mcp-toolbox[all]@latest" stdio`. `pip install "mcp-toolbox[all]"` works too, after which the command is `mcp-toolbox stdio`; `mcp-toolbox sse --host localhost --port 9871` serves SSE instead. Memory data is stored under `~/Documents/zerolab/mcp-toolbox/memory` on macOS, where iCloud syncs it across devices, and `~/.zerolab/mcp-toolbox/memory` elsewhere. The README names \*nix as the main target with Windows expected to work.
One command plus a key — uvx "mcp-toolbox[all]@latest" stdio, then supply credentials