Agent instructions

When an agent connects to a workspace, the server tells it what the workspace is and how to use it. Most MCP clients pass this straight to the model. You can add to it, and if you want, rewrite it entirely. Both settings are under Connect agents, below the setup tabs, and only people who can edit the whole workspace can change them.

Agent instructions

This is the box most teams use. Whatever you write here goes to every connected agent in two places: in the instructions it receives when it connects, and at the top of every get_context result, under "Instructions from the [your organization] team".

It's sent twice on purpose. Not every client shows connection instructions to the model, and some only show them at the start of a conversation. Putting them on every result means they're in front of the agent at the moment it's about to write.

Good things to put here:

• When to use the workspace. "Pull context before writing anything a customer or prospect will read, including short replies."
• Who you're talking to, in a line. "Our customers are clinic managers at independent practices with 2 to 20 staff."
• Standing instructions. "Never mention pricing; send pricing questions to sales@."
• What to do when the context runs out. "If you can't find a fact in the context, say so and suggest who to ask."

Keep it short, up to a few hundred words; the limit is 4,000 characters. Anything long or topic-specific belongs on a page, where it's only sent when it's relevant. Anything that must never be broken belongs on a rule page, where it's labeled as a hard constraint.

Connected agents pick up changes within a minute. Nobody needs to reconnect.

Server instructions

These are the core guidance agents get when they connect: what this server holds, when to call get_context, how to treat rules, how to follow [[Page links]], and how to handle files and private notes. Out of the box, they're written for you, using your organization's name.

You can replace them with your own text (up to 8,000 characters), for example to say exactly which tasks your team uses the workspace for. Reset to default puts the original back.

A few things are always added automatically, whatever you write, so you don't need to keep them up to date yourself:

• a line saying whose context this is and what it covers (built from your page types and most-used tags),
• for sign-in connections, a note that they can reach more than one workspace,
• for connections that can edit, guidance on editing safely,
• your agent instructions.

Most teams never need to touch server instructions. Change your agent instructions first, and only rewrite these if the defaults are actively steering agents wrong.

What agents see

Open What agents see when they connect at the bottom of the section to read the complete text an agent receives: the server instructions with everything added, and the description of the get_context tool. The tool description also names your organization and lists what your context covers, with an example task built from your own tags, so agents can tell when a request is about your company.

It's worth reading this once after you've written a few pages and set up tags, to check it describes your workspace the way you'd describe it.

Instructions or a page?

A rough guide:

If it's...	Put it in...
About when and how to use the workspace	Agent instructions
A constraint on everything that's written	A rule page
Useful for some tasks but not others	A normal page (or a pack)
An instruction about one page, like "only quote prices from here"	That page's usage note