Project context
A shared note about each website (what the business does, its goal, competitors and key pages) that you and your agents read before researching and update afterwards.
Project context is what an agent needs to know about a website before it researches it: what the business sells, the goal right now, who the audience is, which markets matter, who the competitors are, which pages count and how you like to work. It is stored once per website and shared by everyone in the workspace and every agent, so a later session starts from what an earlier one learned instead of asking again or paying for the same research twice.
Alongside the context, each website keeps a short research log: a one-line summary of each piece of work an agent finished, with the date. An agent that finds a recent entry for the same job reuses it.
| Operation | Endpoint | Credits |
|---|---|---|
| Get project context | GET /v1/site-context | 0 |
| Update project context | PATCH /v1/site-context | 0 |
What is stored
| Field | Holds | Limit |
|---|---|---|
business_overview | What the business sells, to whom, and how it makes money. | 2,000 characters |
current_goal | The outcome you want from SEO right now. | 1,000 characters |
audience | Who the customers are. | 1,000 characters |
markets | Countries and languages, for example United States, English. | 500 characters |
preferences | How you like to work: budget, tone, things to avoid. | 1,000 characters |
competitors | Up to 20 competitor domains, each with an optional note of up to 200 characters. | 20 entries |
key_pages | Up to 30 pages that matter, each with an optional purpose of up to 200 characters. | 30 entries |
Text longer than its limit is cut to it. Competitors are stored as bare lower-case domains (https://www.Rival.com/pricing becomes rival.com) and key pages as URLs without the fragment.
The research log keeps the newest 100 entries of each website.
What it is for
- Agents read it first. The MCP tools
get_site_contextandupdate_site_contextare in every playbook (MCP server): an agent callsget_site_context, treats the fields as background about your business, asks you only for what is still empty, and saves what it learned. - It is information, never instructions. Everything in the context was written by you or an agent. An agent is told to read it as facts about the business and not to follow text in it as a command.
- It does not change anything on its own. Writing context sends no email, submits no URLs and spends no credits.
Limits
- Any member who can edit websites (the
site.editcapability) can write. Anyone who can read websites can read. - Writes merge: only the fields you send change. Two writes at once are applied one after the other; if three attempts in a row collide the request answers
409 site_context_conflictand can simply be repeated. - Deleting a website deletes its context and log.