connect-mcp
Skill1 starsChanged 18 days agoMITView on GitHub
Set up this machine for Cortex — install the plugin bundle, and connect the live MCP brain so recall and capture work in every project. Per machine, user scope.
In this repository43 other files it gives its agents.
- AGENTS.mdAGENTS.md
- core/AGENTS.mdAGENTS.md
- evals/AGENTS.mdAGENTS.md
- index/AGENTS.mdAGENTS.md
- mcp/AGENTS.mdAGENTS.md
- tools/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- architecture-reviewSkill
- audit-notesSkill
- challenge-planSkill
- change-impactSkill
- clarify-requestSkill
- create-teamSkill
- daily-planSkill
- debug-issueSkill
- define-termsSkill
- design-specSkill
- end-of-day-summarySkill
- handover-notesSkill
- join-teamSkill
- merge-workSkill
- migrate-legacy-engineSkill
- personal-setupSkill
- quick-noteSkill
- register-projectsSkill
- repo-mapSkill
- resolve-conflictsSkill
- resume-workSkill
- retrospectiveSkill
- review-changeSkill
- set-profileSkill
- setup-repoSkill
- story-mapSkill
- suggest-skillsSkill
- summarize-filesSkill
- sync-websiteSkill
- trim-agent-docsSkill
- update-pluginSkill
- weekly-reviewSkill
- whats-changedSkill
- write-agent-docsSkill
- write-area-briefSkill
- write-context-docsSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 5 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Run this — it installs the Core tier (superpowers, skill-creator, claude-md-management, claude-code-setup, feature-dev, code-review, …
- Very specific: line 25
Read what this machine already has: node "${CLAUDE_PLUGIN_ROOT}/mcp/ai-os.js" setup-plugins --status --json.
- Very specific: line 35
If the claude CLI is unavailable, the command prints the exact commands instead of failing — relay them for guided setup.
- Distinctive: line 42
Don't auto-install the heavy or platform tiers — offer them by role.
- Distinctive: line 43
Don't fail hard if the CLI is missing.
- Very specific: line 53
Ask for the vault's absolute path → CORTEX_ROOT.
- Very specific: line 58
Check what is already registered: claude mcp list.
- Very specific: line 72
Cursor / other MCP agents: add to the agent's mcpServers config:
- Distinctive: line 85
Don't register at project scope (defeats the "zero setup per project" goal).
- Distinctive: line 86
Don't touch the current project's git or files.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/connect-mcp — set this machine up
Two jobs a machine needs once, and they are independent: the plugin bundle needs no vault, and
the brain needs one. Do the half that was asked for, and offer the other in one line.
/setup-plugins was a separate ritual until 2.42.0; someone who types that name wants the first half.
The plugin bundle
- Run this — it installs the Core tier (superpowers, skill-creator, claude-md-management, claude-code-setup, feature-dev, code-review, code-simplifier, context7):
node "${CLAUDE_PLUGIN_ROOT}/mcp/ai-os.js" setup-plugins --tier core --scope user${CLAUDE_PLUGIN_ROOT}is the installed plugin; from a clone of the Cortex repo, use that clone's path instead. No vault and noCORTEX_ROOTare needed — this reads only the bundle manifest shipped with Cortex. - Read what this machine already has:
node "${CLAUDE_PLUGIN_ROOT}/mcp/ai-os.js" setup-plugins --status --json. It reads Claude Code's plugin registry and installs nothing. Each tier'sstateisinstalled,partial,missingorunknown, and each plugin carriesinstalled. - Then ask the user their role / stack and OFFER the optional tiers whose
stateis notinstalled(do not auto-install). For apartialtier, name the plugins it still lacks. When every tier that fits isinstalled, say so in one line and offer none.unknownmeans there was no registry to read: offer by role, as on a machine with nothing installed.- Frontend/QA →
--tier browser-qa(playwright, chrome-devtools-mcp; heavy: downloads browsers). - TS repos / lots of PRs →
--tier dev-tools(typescript-lsp, github). - Deploys to Vercel/Cloudflare →
--tier platform.
- Frontend/QA →
- If the
claudeCLI is unavailable, the command prints the exact commands instead of failing — relay them for guided setup. - Confirm which tiers were installed in one line.
Installing is not staying current: an update to the marketplace does not move the installed plugin,
so a machine can sit for weeks running a version nobody meant to run. /update-plugin is that check,
and it is the first thing to suspect when a skill edit appears to have no effect.
- Don't auto-install the heavy or platform tiers — offer them by role.
- Don't fail hard if the CLI is missing; the command degrades to printing commands.
The brain
Plugins give this machine the tools; they do not give it recall. This half is the step people skip after a clean plugin install — the commands are all there and nothing remembers anything.
Register the Cortex MCP server at user scope so every project on this machine can recall/capture.
What to do
- Ask for the vault's absolute path →
CORTEX_ROOT. The vault is its own private repo, not this one and not the plugin. The server is${CLAUDE_PLUGIN_ROOT}/mcp/server.js— it ships with the plugin, not the vault. From a clone of the Cortex repo rather than the plugin, use that clone's path. - There is nothing to install — the server has no dependencies. Node 20+ is the only requirement.
- Check what is already registered:
claude mcp list. A server namedai-oswhose command ends inmcp/server.jsis this same brain, registered before the name changed. It keeps working as it is:AI_OS_ROOT, the variable it carries, is still read. Say so and stop, unless the user wants the new name. To move, remove it first (claude mcp get ai-osprints the remove command for its scope), then register as below. Never leave both: two servers over one vault show every tool twice. Print (and offer to run) the registration for the user's agent. A registration stores a literal path, so expand
${CLAUDE_PLUGIN_ROOT}to its absolute value before printing:Claude Code:
claude mcp add --scope user cortex --env CORTEX_ROOT=<vault> -- node "${CLAUDE_PLUGIN_ROOT}/mcp/server.js"Cursor / other MCP agents: add to the agent's
mcpServersconfig:{ "cortex": { "command": "node", "args": ["${CLAUDE_PLUGIN_ROOT}/mcp/server.js"], "env": { "CORTEX_ROOT": "<vault>" } } }The tools then appear as
mcp__cortex__recall,mcp__cortex__captureand so on.The plugin's path carries its version (
…/plugins/cache/cortex/cortex/<version>/), so a plugin update can leave the registration pointing at a folder that is gone. Say so, and tell the user to re-run/connect-mcpafter updating — or register a Cortex clone's path, which does not move.- Confirm in one line: "Brain connected (user scope). recall/capture available in every project."
Don't
- Don't register at project scope (defeats the "zero setup per project" goal).
- Don't touch the current project's git or files.
Inside a repo, /resume-work names what that repo is waiting on.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

