# AI Agents for Strategy Execution | Vindaris

> Connect Claude, ChatGPT, or Claude Code to Vindaris over MCP. An agent reads
> execution risk from work activity, drafts check-ins and summaries, and creates
> or updates goals, KPIs, and tasks, all within the permissions you set. A human
> approves what enters the plan.

This is the Markdown version of https://vindaris.com/ai-agents, for AI agents.

## What Vindaris is

Strategy execution software. You map strategy, goals, KPIs, projects, and tasks,
then integrate the tools where work happens. The Work Graph connects it all and
flags execution risk, the gap between a goal and the effort actually pointed at
it, read from work activity rather than a typed status.

## Connect over MCP

- MCP server (Streamable HTTP): `https://app.vindaris.com/api/mcp`
- REST alternative (OpenAPI): `https://app.vindaris.com/api/ai/openapi`
- Auth: OAuth 2.0 with dynamic client registration. Protected-resource metadata
  at `https://app.vindaris.com/.well-known/oauth-protected-resource`. A personal
  token also works as `Authorization: Bearer mcp_...`.
- An org admin turns on AI Tools and grants access before an agent can act.

### Claude Code

```
claude mcp add vindaris \
  --transport streamable-http https://app.vindaris.com/api/mcp \
  --header "Authorization: Bearer mcp_YOUR_TOKEN"
```

### Project config (.mcp.json)

```json
{ "mcpServers": { "vindaris": { "type": "http", "url": "https://app.vindaris.com/api/mcp" } } }
```

### Claude Desktop, ChatGPT, Codex

Add the server URL as a connector. OAuth handles the sign-in; no token is pasted
into the model.

## Permissions

Access is bounded per workspace. An admin sets each resource (strategy, goal,
kpi, project, task) to one of `off`, `read`, `write`, `read_write`, and toggles
individual tools. The default is `off`. An agent can be kept read-only or allowed
to act, and the actions that change the plan can require human approval.

## Tools

Read:
`list_goals`, `get_goal`, `list_kpis`, `get_kpi`, `list_projects`, `get_project`,
`list_strategys`, `get_strategy`, `list_tasks`, `get_task`

Create and update:
`create_goal`, `update_goal`, `create_kpi`, `update_kpi`, `create_project`,
`update_project`, `create_strategy`, `update_strategy`, `create_task`, `update_task`

Risk and summary:
- `get_work_graph_alerts` - prioritized items needing attention: at-risk goals,
  stale tasks, pace warnings, alignment gaps.
- `generate_summary` - structured read of the whole workspace with status and
  alignment.

Map and annotate:
- `align_entities` / `remove_alignment` - connect or disconnect work and objectives.
- `push_comment` - comment on a task; syncs to Jira, Slack, or Asana.

Other:
- `ingest_content` - turn notes or an email into aligned tasks, goals, or projects.
- `trigger_reminder` - nudge the owner of an overdue item.
- `toggle_risk_bell` - subscribe a Data Studio, dashboard, or heatmap view to risk alerts.

A machine-readable manifest of all tools is embedded in the HTML page as
`<script type="application/json" id="vindaris-mcp">`.

## Example session

```
Human: Where is the European expansion strategy slipping, and draft the weekly update.

Agent:
  -> get_work_graph_alerts()
  <- "Make onboarding self-serve" at risk, Activation rate at 38% of target, 2 stale tasks
  -> generate_summary({ strategy: "Expand into the European enterprise market" })
  <- draft returned
  -> create_task({ title: "Unblock self-serve onboarding", align: "goal" })
  <- task created, pending human approval

  The expansion is on track except self-serve onboarding, where the Activation
  rate sits at 38% of target and two tasks have not moved in nine days. I drafted
  the weekly update and opened a follow-up task. Both wait for your approval.
```

Example only. The tool names and the human-approval step are real; the data is illustrative.

## FAQ

**How do AI agents connect to Vindaris?** An admin turns on AI Tools and connects
an assistant (Claude, ChatGPT, Claude Code) over MCP at
`https://app.vindaris.com/api/mcp`. Sign-in is OAuth, so no API key is pasted into
the model.

**What can an agent do?** Read Work Graph alerts and work activity, generate AI
summaries, create or update goals, KPIs, projects, and tasks, align items, and
comment, all within the permissions it is granted.

**Agent Skills vs the MCP connection.** The downloadable Agent Skills
(https://vindaris.com/skills) help an assistant draft the plan in any tool. The
MCP connection is how an agent then acts on a live workspace. The skills draft;
MCP runs it.

**Can I control what an agent changes?** Yes. Fine-grained permissions decide what
an agent can read, change, and what needs human approval.

## Links

- Overview (HTML): https://vindaris.com/ai-agents
- Connect guide: https://vindaris.com/help/connect-ai-assistants-mcp
- Fine-grained permissions: https://vindaris.com/help/fine-grained-permissions
- Agent skills: https://vindaris.com/skills
- Site index for LLMs: https://vindaris.com/llms.txt