CmdBrief MCP server for agents and assistants
A read-only MCP server that answers questions about CmdBrief, its guides, and the /compare tool data — served as one dependency-free file from cmdbrief.com.
If you are asking a coding agent whether CmdBrief fits your setup, it should not have to scrape a marketing page to find out. This server hands it the answer directly: the guides, the comparison data, and CmdBrief’s own product facts, from the same files this site renders.
It is read-only. It has no authentication, stores nothing, and cannot change anything — on your machine or ours.
The short answer
One file, no dependencies, Node 20 or newer. Download it, point your MCP client at it, and your agent can ask what a CmdBrief guide says, what Warp costs today, or which tool documents usage windows — with the source URL and check date attached to every value.
This is not CmdBrief’s in-app MCP mode, which is a different thing: the app can act as an MCP server for your own agents against your own workspaces. This one only knows about the public website.
Install it
- Download the server. It is a single file and it stays current on its own, because it reads its data from this site.
- Register it with your MCP client.
- Ask your agent something — “what does the CmdBrief comparison say about Ghostty?” is a good first question.
curl -fsSL https://cmdbrief.com/mcp/server.mjs -o ~/.cmdbrief-mcp.mjs
claude mcp add cmdbrief -- node ~/.cmdbrief-mcp.mjsFor a client configured through a JSON file, the same server looks like this:
{
"mcpServers": {
"cmdbrief": {
"command": "node",
"args": ["/Users/you/.cmdbrief-mcp.mjs"]
}
}
}What it exposes
| Tool | What it does | Arguments | Returns |
|---|---|---|---|
list_guides | List every CmdBrief guide: slug, title, description, category, and the date it was last updated. | None | One entry per guide, with its page URL and its Markdown mirror URL. |
get_guide | Fetch one CmdBrief guide or comparison page in full, as Markdown. | slug | The same Markdown the page serves at its .md mirror. |
list_comparisons | List the tools CmdBrief is compared against, with the date each comparison was last checked. | None | One entry per comparison page, with the compared tool’s current version and check date. |
get_compare_data | Get the structured comparison data for one tool: version, release date, platforms, pricing, licence, and a capability verdict per row — each with the source it came from. | tool — one of warp, ghostty, iterm, cursor, claude-desktop | The tool’s facts and capability cells beside CmdBrief’s, with a source URL and check date on every value. |
get_cmdbrief_facts | Get CmdBrief’s own product facts: supported agent CLIs, platform requirements, current version, pricing, and refund terms. | None | The same facts the pricing and features pages render, from the same data file. |
Five tools, all reads. Every comparison value comes back with the source URL it was taken from and the date it was last checked, so an assistant can cite it rather than paraphrase it.
Where the data comes from
The same place the pages get it. Guide and comparison text is generated from the site’s content files, which is also what the Markdown mirrors serve. Tool versions, release dates, and licences are pulled from each vendor’s own feed — a Homebrew cask, a GitHub tag, a Sparkle appcast, an update endpoint — on a scheduled refresh. Pricing, platform support, and capability verdicts are read by a person and carry their own date.
- Nothing is written by a model. Every value is either pulled from a vendor feed or typed by a person against the vendor’s page.
- A capability we could not find documented reads as “—”, never as “no”. The comparison says what a vendor documents, not what we assume.
- Every value is dated. If a check is old, you can see that it is old.
The rendered version of the same data is on the comparison pages, and the guides it indexes are on the guides hub.
What it does not do
- No writes. There is no tool that changes anything.
- No authentication and no accounts. It never asks for a key and never sends one.
- No state. It keeps nothing between calls and writes nothing to disk.
- No telemetry. It does not report what you asked it.
- Not a support channel. For anything the data cannot answer, email [email protected].
No. It is read-only public data, with no authentication of any kind. You do not need a CmdBrief subscription to run it.
The server reads its data from cmdbrief.com at startup, so it is as current as the site. Every value carries the date it was last checked, and the comparison pages print the same date.
No. CmdBrief the app can act as an MCP server for your own agents against your own workspaces, with in-app approval for staged actions. This server only answers questions about the public website.
It needs one HTTPS request at startup to fetch its data. If you place a data.json next to the server file, it reads that instead and makes no network call at all.
Primary sources
Provider terms and product behavior can change. Confirm plan, pricing, model, and permission details in the current official documentation.