# What is OpenNomos

> A growth capability layer for your own agent, with project memory that outlives any one session.

URL: https://mcp-docs.opennomos.com/docs

OpenNomos is **not an agent**. It is a layer your own agent (Claude Code, Cursor, Codex, or any MCP client) connects to, so that agent can do growth work without you wiring up a dozen scrapers and APIs, and without forgetting everything when the session ends.

It gives the agent two things:

1. **Capabilities.** Tools that read public sources and return the data: conversations on the web, Reddit, X, Xiaohongshu, Zhihu and Weibo, SEO and keyword data, what ChatGPT answers about your category, app reviews and video comments, creator profiles, competitor tech stacks and ads, news and trending lists, and any page as clean markdown. Results are returned to the agent. They are never saved on their own.
2. **Project memory.** A server-side record, per project, that the agent writes **explicitly**: free-form notes (decisions, rules, drafts, conclusions), plus an optional follow-up state for the people and posts worth acting on (opportunities), small experiments with a deadline, short links, and the actions you took.

<Cards>
  <Card title="Quickstart" href="/docs/start/quickstart">
    Connect over MCP, authorize, and verify with `discover_tools`.
  </Card>
  <Card title="Core concepts" href="/docs/start/concepts">
    Project, memory, opportunity, experiment, action and briefing in one page.
  </Card>
  <Card title="The growth loop" href="/docs/growth-loop">
    Pull a briefing, work, write back with `track` and `log_action`.
  </Card>
  <Card title="Tools reference" href="/docs/tools">
    All 21 tools: parameters, examples, errors and cost.
  </Card>
</Cards>

## What OpenNomos deliberately does not do

- **It does not post, reply or message anyone.** Your agent or you act elsewhere, then record it. The tools only record.
- **It does not decide.** No fixed workflow, scoring of your ideas, categories or required order. The growth tools are optional and can be used one at a time.
- **It does not infer.** Stages are never guessed from actions, and results with no known source are counted as `unknown`, not credited to a channel.
- **It does not save by itself.** Memory is written only when the agent asks.

## Connection points

| | |
| --- | --- |
| MCP (Streamable HTTP, OAuth) | `https://mcp.opennomos.com/mcp` |
| REST | `https://service.opennomos.com/v1/tools` |
| Agent setup instructions | `https://service.opennomos.com/SKILL.md` |
| Console (accounts, keys) | `https://tools.opennomos.com/console` |

## For agents reading this site

Every page has a markdown twin at `/docs/<page>.md`, the full text is at [`/llms-full.txt`](https://mcp-docs.opennomos.com/llms-full.txt), and the index is at [`/llms.txt`](https://mcp-docs.opennomos.com/llms.txt).