Ezeo inside your AI coding agent
Two ways to give Claude Code, Codex, Cursor, Gemini CLI and other agents your Ezeo data: the remote connector, or the local ezeo mcp server.
Pick a route
Both routes put Ezeo tools inside your agent. They differ in who can use them and what they can do.
| Remote connector | Local server (ezeo mcp) | |
|---|---|---|
| Who | Remote connectorAnyone with an Ezeo login. | Local server (ezeo mcp)Owners, admins and members of an Ezeo organization. Client accounts are refused. |
| Install | Remote connectorNothing to install: one command or config entry pointing at the Ezeo URL. | Local server (ezeo mcp)npm i -g @ezeo/cli, then ezeo login and ezeo setup. |
| Sign-in | Remote connectorOAuth in your agent, approved on an Ezeo page. | Local server (ezeo mcp)ezeo login, stored in a file on your machine. |
| Tools | Remote connector16 tools: read-only, plus request_articles with a preview and a confirmation code. | Local server (ezeo mcp)33 tools covering research, audits, reports and approvals, each paid one behind a confirm token. |
| Status | Remote connectorVerified in Claude Code, Codex and Cursor (as of 2026-10-15). Gemini CLI not verified: personal Google sign-in is no longer supported. | Local server (ezeo mcp)Browser sign-in verified live on 2026-09-30; other commands tested against recorded responses. |
Route 1: the remote connector
For any agent that supports remote MCP servers with sign-in. These are the commands and config entries from the connect guide: add one, then sign in from the agent.
| Agent | Command or config |
|---|---|
| AgentCursor | Command or config~/.cursor/mcp.json{
"mcpServers": {
"ezeo": {
"url": "https://ezeo.ai/mcp"
}
}
} |
| AgentGitHub Copilot | Command or config.vscode/mcp.json{
"servers": {
"ezeo": {
"type": "http",
"url": "https://ezeo.ai/mcp"
}
}
} |
| AgentWindsurf | Command or config~/.codeium/windsurf/mcp_config.json{
"mcpServers": {
"ezeo": {
"serverUrl": "https://ezeo.ai/mcp"
}
}
} |
| AgentClaude Code | Command or configCommandclaude mcp add --transport http ezeo https://ezeo.ai/mcp |
| AgentOpenAI Codex | Command or config~/.codex/config.toml[mcp_servers.ezeo]
url = "https://ezeo.ai/mcp" |
| AgentGemini CLI | Command or configCommandgemini mcp add --transport http ezeo https://ezeo.ai/mcp |
Cursor and GitHub Copilot also have a one-click Add button on the connect page. Gemini CLI no longer accepts personal Google accounts, so it needs a Gemini API key or Vertex AI.
Step-by-step setup for every assistantRoute 2: the local server
ezeo mcp runs an MCP server on your machine over stdio and serves the CLI's commands to your agent as tools. ezeo setup registers it for you. Install the CLI from npm (Node 22 or newer), sign in, then set up your agent:
npm i -g @ezeo/cli
ezeo login
ezeo setup claude --dry-run
ezeo setup claudeThese are the agents ezeo setup supports and exactly what it does for each. It prints the line before it runs it.
| Agent | Command | What it runs or writes |
|---|---|---|
| AgentClaude Code | Commandezeo setup claude | What it runs or writesclaude mcp add --scope user ezeo -- ezeo mcp |
| AgentOpenAI Codex | Commandezeo setup codex | What it runs or writescodex mcp add ezeo -- ezeo mcp |
| AgentGrok | Commandezeo setup grok | What it runs or writesgrok mcp add --scope user ezeo -- ezeo mcp |
| AgentGemini CLI | Commandezeo setup gemini | What it runs or writesgemini mcp add --scope user ezeo ezeo mcp |
| AgentCursor | Commandezeo setup cursor | What it runs or writesmerge {"mcpServers":{"ezeo":{"command":"ezeo","args":["mcp"]}}} into ~/.cursor/mcp.json |
- The CLI and the connector both register an MCP server named ezeo, so use one per agent: the connector for most people, the CLI for terminal power users.
- Sign in first with ezeo login. The server uses that session, so the agent acts as you.
- Add --dry-run to print what it would do without changing anything.
- Running it twice is safe: an existing ezeo entry is reported as already configured.
- For Cursor it merges one entry into ~/.cursor/mcp.json and keeps your other servers. If the file is not valid JSON it stops rather than overwrite it.
- When ezeo is not on your PATH (a build from source), the launcher it registers is node <path to ezeo-cli>/dist/index.js mcp instead of ezeo mcp.
Your agent then sees one tool per command, named ezeo_ followed by the command, for example ezeo_articles_waiting. Paid tools return a dry run and a confirm token first.
Every local tool and its costTeach the agent the rules
ezeo help --agent prints the agent reference: the rules on unmeasured values, paid tools and article approval. ezeo skill install writes it as a Claude skill.
ezeo help --agent
ezeo skill install --dry-run
ezeo skill installThe skill is written to ~/.claude/skills/ezeo/SKILL.md, a Claude location. For other agents, ezeo help --agent prints the same text to add to their instructions.
Example prompts
Try these once connected. Name one of your own projects where it helps.
- Connector What needs my attention in Ezeo?
- Connector Give me this week's SEO review for my main project.
- Connector Plan my next three articles from Ezeo's content ideas, and show me the preview before you request anything.
- Local server Use ezeo_articles_waiting and tell me which article runs need me.
- Local server Check ezeo_status for my main project: is any data source stale?
- Local server Show me ezeo_keywords_opportunities and draft a plan. Do not run any paid tool without asking me.
On the connector, five ready-made prompts such as weekly_seo_review also appear as slash commands in agents that list MCP prompts. They are on the MCP tools page.