Getting startedAdd a server, check it works, install it
Run toolshed and you get a list of your tools, plus a few built-in
examples. The keys:
| Key | |
|---|---|
↑ ↓ | move |
a | add a tool |
t | start the server and list the tools it offers |
i / enter | install (or approve, for a tool your agent added) |
u | uninstall |
d | delete one of your tools |
m | import the servers Claude Code and Codex already have |
q | quit |
Adding a tool asks what kind it is (an npm or Python package, a command, or a remote URL) and what to call it. toolshed then starts it and lists its tools, so you know it works before your agent tries it. Installing asks for any values the tool needs and where to install it: for every project, or just this repository.
From the command line
toolshed add memory --npx @modelcontextprotocol/server-memory
# ✓ Added memory
# ✓ pinned to 2.1.0
# ✓ memory offers 9 tools
# create_entities Create multiple new entities in the knowledge graph
# …
toolshed install memory
# ✓ memory → claude, codex
More kinds of tools:
toolshed add fetch --uvx mcp-server-fetch # Python package
toolshed add docs --url https://mcp.context7.com/mcp # remote server
toolshed add jira --secret JIRA_TOKEN -- jira-mcp --site acme # any command, with a secret
toolshed add acme --url https://mcp.acme.com/mcp \
--header 'Authorization=Bearer ${ACME_TOKEN}' --secret ACME_TOKEN # remote server with a token
Secrets are asked for at install, at a hidden prompt, and go to your keychain (see Secrets). npm and Python packages are pinned to their current release (see Updates and versions).
Next
- Already have servers set up by hand? Import them.
- Let your agent add servers itself: Your agent.
- Share a set of tools with your team: The registry.