What the agent can do with it
For anything the skill does not cover, the agent reads these docs through Meridian MCP (
docs_search, docs_get) rather than answering from memory.
What stays with you
The skill tells the agent to leave these to you, explain why, and wait:- Signing in. The agent asks you to run
meridian loginin your own terminal, then checks the result withmeridian whoami. - Approving Meridian MCP on the consent screen in your browser.
- API keys and secrets. You create
MERIDIAN_API_KEYon the app’s Settings → SDK page and put it in.envor your hosting variables yourself. The agent never asks you to paste a key, password or secret into the chat. - Dashboard setup: connecting GitHub, adding a database, setting secrets, adding a domain, choosing a plan.
Install it
Pick one. All three install the same skill.- Meridian CLI
- Claude Code plugin
- Codex plugin
- skills CLI
From your app’s root:The command writes the skill for each agent the project already uses (all three when it finds none) and adds Meridian MCP to that agent’s project config:
It merges with config files that already exist and never overwrites another server. Commit these folders so your teammates’ agents get the skill too.
Run
meridian skills install again after upgrading the CLI to update the skill. Replacing a skill folder that changed asks first; --yes approves it.Connect Meridian MCP
The first time the agent calls Meridian MCP, approve the connection:Try it
Once the skill is installed, ask your agent for a task in your own words, for example:- “Add Meridian billing to this app, with a pricing page and a gate on advanced reports.”
- “Get this app running locally on my Meridian database.”
- “Deploy main to production and tell me if anything fails.”