Build with an AI agent

Point Claude, Cursor, or Codex at Omi’s llms.txt, MCP server, and Developer API to build on your own data.

Three things let a coding agent build on Omi:

llms.txt

This whole site, indexed for LLMs. Paste it into any agent.

Omi MCP server

Give an agent live access to your memories and conversations

Developer API

REST endpoints the agent can call with your API key

Copy-paste prompts

Explore the platform

Read https://docs.omi.me/llms.txt. Summarize the ways I can extend Omi
(apps, webhooks, chat tools), then list the first three endpoints I would
need to list and create memories over the Developer API, with curl
examples using a Developer API key (omi_dev_...).

Build an app

Read https://docs.omi.me/llms.txt, especially the "Apps" pages. Design a
small integration app that receives conversation-finished webhooks and
posts a daily summary to my own server. Give me the app configuration,
the webhook payload handling in Python, and the review checklist I must
pass before submitting it in the Omi app.

Query my data

You are connected to my Omi account over MCP. Search my memories and
conversations from the last 7 days for anything about the renewal date
of my domain, then draft a reminder with links back to the exact
conversations you used.

Giving an agent your data

The MCP server exposes read and write tools scoped to your account: memories, conversations, action items, goals, and people. Set it up once in your MCP client (Claude Desktop, Cursor, and anything else that speaks MCP), and the agent can query your data directly:

claude mcp add --transport http omi https://api.omi.me/v1/mcp

See MCP setup for OAuth, scoped API keys, and client-specific snippets.

Treat an agent with Omi access like a person with your calendar and your call log. Prefer read-only scopes for anything you do not fully control, and rotate keys from the app when you are done.