Add the server to your crushrc
Crush supports MCP servers through three transport types: stdio for command-line servers, http, and sse, per the Crush README. The KnowMan MCP server is a standalone stdio server, started with npx -y @knowman/mcp on Node 22 or newer, per the MCP server guide, so the local stdio form is the one to use.
Crush reads crushrc files, which are Bash with Crush-specific builtins, per the Crush README. On Unix-like systems it checks ./.crushrc, then ./crushrc, then ~/.config/crush/crushrc, with the same order under Windows paths. mcp add registers a local server, --command names the executable, --args supplies arguments, and --env passes environment variables. The snippet below applies these documented options to KnowMan's documented start command.
Create a KnowMan API key
In the KnowMan web app, open Settings > Connections > Connect an AI agent to create a key. Keys look like km_..., and the server uses the key as a Bearer key against KnowMan's HTTP API through the environment variable KNOWMAN_API_KEY, per the MCP server guide. Set KNOWMAN_API_KEY in your shell before launching Crush, so the --env entry in the snippet passes it to the server.
Sign in once with knowman login
KnowMan documents an alternative to the key in the config: run npx -p @knowman/mcp knowman login once, sign in in the browser, and the saved credential is shared by the CLI and the MCP server, so the env entry can be dropped, per the MCP server guide.
The same package installs the knowman CLI, which includes login, status and logout commands, per the MCP server guide. If you prefer to keep the key route, leave the --env KNOWMAN_API_KEY entry in place.
Verify the setup
The Crush README does not document a way to verify that an MCP server connected, per the Crush README. What you can check is the credential side: the knowman CLI includes a status command, per the MCP server guide. The KnowMan server exposes kb_recent, which lists recently saved items, among its tools, per the MCP server guide.
The KnowMan tools
The KnowMan MCP server exposes these tools, per the MCP server guide. kb_search searches saved items by keyword or meaning and returns titles, URLs and metadata. kb_ask answers a question with citations. kb_get_item returns the readable content of one item.
kb_save saves a URL to the Inbox or the Knowledge Base with an optional note. kb_recent lists recent items. kb_memories lists saved claims, decisions and playbooks, per the MCP server guide.
Common problems
Config location. On Unix-like systems Crush checks ./.crushrc, then ./crushrc, then ~/.config/crush/crushrc, and the same order applies under Windows paths, per the Crush README. Crush respects XDG_CONFIG_HOME, so your paths may differ, and CRUSH_GLOBAL_CONFIG can override the user config location.
Old JSON config. The legacy JSON format is still supported but deprecated, and any $(...) in crush.json runs at load time, per the Crush README. Application state is separate from configuration and should not be edited by hand, per the Crush README.
Review before you launch. A crushrc is trusted code that runs in a full shell, and the README advises not launching Crush in a directory whose config you have not reviewed, per the Crush README.
What KnowMan brings to Crush
KnowMan saves articles and links, reads and tags them, answers questions with citations, and keeps memories: claims, decisions and playbooks, per the personal knowledge base guide. Through the server's tools, a client running the server can search that library, ask it questions, and save new URLs into it. For the wider setup story, see an MCP server for a knowledge base and a knowledge base for AI agents.
Questions people ask
How do I add an MCP server to Crush?
Add an mcp add entry to your crushrc, which is Bash with Crush-specific builtins. Use --command for the executable, --args for its arguments, and --env for environment variables. The syntax is documented in the Crush README.
Where does Crush read its config from?
On Unix-like systems it checks ./.crushrc, then ./crushrc, then ~/.config/crush/crushrc, with the same order under Windows paths. Crush respects XDG_CONFIG_HOME, so your paths may differ. See the Crush README.
Does the KnowMan MCP server work with Crush?
Crush documents support for stdio MCP servers alongside http and sse, per the Crush README, and the KnowMan server is a standalone stdio server started with npx -y @knowman/mcp on Node 22 or newer, per the MCP server guide. The setup on this page combines both projects' documented formats.
Do I need an API key in the config?
You need one of the two documented routes. Create a key at Settings > Connections > Connect an AI agent and pass it as KNOWMAN_API_KEY, or run npx -p @knowman/mcp knowman login once, after which KnowMan documents that the env entry can be dropped; both routes are covered in the MCP server guide.