Most people start by pasting. They copy the annual goals into the chat, or attach the strategy deck, and ask the agent to help plan the week. It works once. By the second week two of the goals have moved and the paste is seven days old, so the agent plans, with great confidence, against a company that no longer exists in quite that form.
There are four ways to give an agent context on your goals, and they differ mainly in who has to keep that context current.
What are the ways to give an agent context on goals?
- Paste it into the session. You keep it current, by hand, every time. This suits a one-off question and nothing else.
- Write an instruction file. A file such as AGENTS.md or a project brief loads automatically, which removes the pasting. However, someone still has to edit it when the plan changes, and in practice nobody does. We describe how these files age in Why AI Agents Get Your Company Context Wrong.
- Connect the tools. With MCP servers for Jira and Slack, the agent can look things up itself. The information is current, but it arrives as raw tickets and messages. On every run the agent has to work out again which ticket serves which goal, and it will not always work it out the same way. What an MCP Server Gives an AI Agent covers that limit.
- Connect a context layer. One system holds the plan and the activity already joined, and the agent queries it. The upkeep moves from a person to the system, which follows the activity in your tools.
The fourth is the only one in which nobody maintains anything by hand, and for that reason it is the only one that survives a busy quarter.
What does an agent need to know about a goal?
Whichever route you choose, it helps to be precise about what "context on a goal" means. An agent can be useful once it knows four things. It needs to know what the goal is and who owns it, with the target and the due date.
It needs to know what serves the goal, meaning the KPIs that measure it and the projects and tasks meant to move it, because without those links the agent cannot tell whether a given task matters.
It needs the state of the goal according to the evidence: a typed status says what the owner believed when they last looked, whereas stalled tickets and a blocker mentioned in a meeting say what is happening now. And it needs to know what changed recently, since a due date pushed twice or an owner replaced last week changes how it should weigh everything else.
A slide deck contains the first of these. The other three live in your work tools and change daily.
How do you connect an agent to your goals over MCP?
Vindaris holds all four in its Work Graph. It maps the strategy to the goals, KPIs, projects and tasks under it, and it analyzes work activity across the tools your teams use, such as Jira, Slack, Microsoft 365 and meeting transcripts. An agent reaches the graph through the Vindaris MCP server.
Setup takes two steps. An admin turns on AI Tools in Vindaris and sets each resource (strategy, goal, KPI, project, task) to off, read, or read and write; the default is off. Then you add the server to your agent. In Claude Desktop, ChatGPT or Codex you add https://app.vindaris.com/api/mcp as a connector and sign in with OAuth, so no key is pasted into the model. In a project that uses Claude Code, one line in .mcp.json does it:
{ "mcpServers": { "vindaris": { "type": "http", "url": "https://app.vindaris.com/api/mcp" } } }
Now ask a question that needs the plan and the activity together: "Which projects under our onboarding goal are stuck, and since when?" The agent calls list_goals to find the goal, get_goal to get the projects and KPIs linked to it, and get_work_graph_alerts to see which of them are blocked or have gone quiet. It can answer with the project and the date the blocker appeared, and give the reason, weeks before the quarterly review would have raised it.
More examples of this kind, with the tools each one calls, are in the public Work Graph schema repository. The connection details for each client are on the AI agents page, and the concept behind all of it is in What Is an AI Context Layer?.
Where should you start?
You do not need every agent connected on the first day. Pick the question your leadership team asks most often and answers least reliably (for many it is some form of "are we on track?") and check what your agent says today against what the tickets say. The distance between those two answers tells you how much of your AI's advice rests on context that someone pasted in and forgot.
And if the distance turns out to be large, how many decisions has the agent already informed?
FAQ
Can I just upload our strategy document to the AI? You can, and it helps for a one-off question. The document goes out of date as soon as a goal or a project changes, and it says nothing about which tasks are moving.
Which AI agents can connect to Vindaris? Any MCP client, including Claude, Claude Code, ChatGPT and Codex. Sign-in runs over OAuth, so nobody pastes an API key into the model.
Can an agent change our goals? Only if an admin allows it. Each resource is set to off, read, or read and write, with off as the default, and changes to the plan can require a person's approval.
Do we need to restructure our goals first? No. Vindaris maps the strategy, goals, KPIs, projects and tasks you already have and links tasks from tools such as Jira and Asana to them.