All servers / Other / B3OS MCP server
B3OS MCP server
B3OS MCP server over stdio, and the shared MCP tool table both B3OS MCP servers serve: vision, render, infer, sandboxes, routines and actions as tools.
Tested by ConnectATDScore 80 out of 100. Tested 2026-11-19 on package version 0.2.1. Package: @b3os/mcp · source
Score breakdown
| Runs | 18 / 20 | |
|---|---|---|
| Tool descriptions | 14 / 20 | |
| Package safety | 20 / 20 | |
| Maintenance | 20 / 20 | |
| Adoption | 8 / 20 |
What we saw
- Starts and answers the MCP handshake: yes (1.6 seconds including download)
- Tools offered: 15
- Last published: 2026-11-15
- Weekly downloads (npm): 536
- License: Apache-2.0 · Direct dependencies: 1
Want this server set up for your team?
ConnectATD builds AI automation and wires up MCP servers like this one. Tell us what you are trying to do.
Talk to ConnectATDTools and how they are described
Agents choose tools from these descriptions. Under each name you can see how many of its parameters are explained.
vision_scrape1 of 12 parameters describedFetch one URL and return its main content as Markdown (or html, text, links, a screenshot, or JSON filled from a schema by the node's model). Answers from the cache when a fresh copy exists, uses a plain HTTP fetch when that is enough, and a real browser when the page needs JavaScript.
vision_map0 of 6 parameters describedDiscover the URLs of a site from its links and sitemaps, optionally ordered by relevance to a search phrase.
vision_crawl2 of 11 parameters describedCrawl a site breadth first from a URL (links and sitemaps, at most 500 pages) and return the pages as Markdown. Waits up to wait_s seconds; if the crawl is still running, call again with job_id.
vision_act1 of 6 parameters describedRun browser actions in a real Chrome session in one call (navigate, click, type, key, scroll, select, wait, evaluate, screenshot, markdown) and return each action's result. Without session_id a session is created for this call and closed after it; keep_session: true keeps it and returns its id for the next call.
render_screenshot4 of 11 parameters describedCapture a page (or one element, or the full page) in a warm browser, optionally styled with a background and window frame. Give url or html. Returns the image and a signed URL to it.
render_video5 of 10 parameters describedRecord a scripted product demo (beats of browser steps, optional narration and captions) as MP4, GIF and a poster image. Waits up to wait_s seconds and returns the poster; if the video is still rendering, call again with job_id.
sandbox_create4 of 9 parameters describedStart a sandbox: an isolated Firecracker microVM (the base template has Ubuntu 24.04, Python 3.12, Node 24 and git). It stops after timeout_s seconds. Use sandbox_exec, sandbox_read_file and sandbox_write_file on its id.
sandbox_exec1 of 8 parameters describedRun a shell command (a string, run with /bin/sh -c) or an argv array in a sandbox and return its exit code, stdout and stderr. background: true returns at once with the process.
sandbox_read_file0 of 4 parameters describedRead a file from a sandbox (UTF-8 text, or base64 for binary), or list a directory with list: true.
sandbox_write_file0 of 5 parameters describedWrite a file in a sandbox, replacing it if it exists; parent directories are created.
sandbox_delete0 of 1 parameter describedStop a sandbox and destroy its memory and disk. Its snapshots are kept.
routine_run1 of 5 parameters describedStart a run of a published routine (by name or rtn_ id) with an input object and wait up to wait_s seconds for its output; if it is still running, call again with job_id.
actions_search0 of 5 parameters describedFind catalog actions (GitHub, Slack, Linear, Notion, HTTP and more) by what they do. Returns refs to pass to actions_describe (for the input schema) and actions_call. kind=trigger finds triggers for routines instead.
actions_describe0 of 1 parameter describedGet an action's input and output JSON Schemas, effect class and the connections it needs.
actions_call2 of 4 parameters describedRun a catalog action with an input object. Credentials come from your stored connections (vault), never from arguments; name one per connection key in connections when you have several.
Tested in a sandbox with an empty environment: no API keys, a throwaway folder, nothing from a real account. A passing test shows the server runs and responds. It is not a security audit.