StellaryStellary
FeaturesHow It WorksWhy StellaryBlog
Overview
Concepts & architecture
Getting Started
Your first project in 5 min
API Reference
Complete REST API docs
MCP Integration
Connect AI agents
FAQ
Sign inStart Free
FeaturesHow It WorksWhy StellaryBlog
Documentation
Overview
Concepts & architecture
Getting Started
Your first project in 5 min
API Reference
Complete REST API docs
MCP Integration
Connect AI agents
?
FAQ
Sign inStart Free
StellaryStellary

The AI-powered command center for teams that ship.

Product

  • Features
  • How It Works
  • Why Stellary
  • Blog
  • FAQ

Developers

  • Documentation
  • API Reference
  • MCP Integration
  • Getting Started

Company

  • FAQ
  • Legal Notice
  • Terms of Service
  • Privacy Policy
  • Cookie Policy
  • DPA

© 2026 Stellary. All rights reserved.

Legal NoticeTerms of ServicePrivacy PolicyCookie PolicyDPA
Overview
Guide
  • User Guide
  • Board & Cards
  • Knowledge Base
  • Cockpit & Command Center
  • AI Project Wizard
  • AI Agents & MCP
  • Automations
  • Team & Collaboration
Developers
  • Getting Started
  • API Reference
  • MCP Integration
AI Agents & MCP

AI Agents & MCP

Deploy intelligent agents inside your workspace. They can read your board, create and update cards, manage documents, and collaborate with your team — all with configurable autonomy and full audit trails.

What are AI Agents?

AI Agents in Stellary are workspace-scoped bot users that can interact with your projects using a curated set of tools. Unlike a simple chatbot, agents operate on real data — they create cards, move tasks, write documents, and post comments just like a human team member would.

Each agent has its own identity, permissions, and tool configuration. You decide what an agent can do, how much autonomy it has, and how many actions it can perform within a given time window.

Creating an Agent

Navigate to Settings → AI Agents in your workspace. Click "New Agent" and configure:

  • Name & avatar — give your agent a recognizable identity
  • Autonomy mode — choose between Supervised, Autonomous, or Approval
  • Tools — select which tool categories the agent can access
  • Rate limits — define the maximum number of actions per time window

Once created, the agent appears in your workspace member list and can be assigned missions immediately.

Autonomy Modes

Autonomy modes control how much freedom an agent has when executing actions. Pick the level that matches your trust and risk tolerance:

Supervised

The agent proposes changes but never executes them directly. A human must approve or reject each proposal before anything happens.

Autonomous

The agent executes actions directly without waiting for approval. Best suited for routine, low-risk operations where speed matters.

Approval

The agent batches its proposed changes for review. A team member can approve or reject the entire batch at once.

Agent Tools

Tools are the actions an agent can perform. They are organized into categories, and you enable or disable entire categories when configuring an agent. Every tool call goes through a policy check before execution and is logged in the audit trail.

Board Read

List projects, columns, and cards. Query card details, priorities, assignees, and statuses across your workspace.

Board Write

Create, move, and update cards. Change statuses, set priorities, edit descriptions, and manage columns.

Board Collaboration

Assign members to cards, post comments, and manage labels. Automate team coordination workflows.

Document Operations

List, read, create, and update documents in the knowledge base. Agents can analyze and produce documentation autonomously.

Missions

A mission is a task you assign to an agent at the card level. Open any card, go to the AI tab, and describe what you want the agent to do — for example, "Break this feature into subtasks with priorities" or "Draft a technical specification document."

Once launched, the mission streams its progress in real time using Server-Sent Events (SSE). You can watch tool calls, intermediate results, and the final output as it happens — no need to refresh.

Mission Lifecycle

Every mission moves through a well-defined lifecycle. The current stage is displayed in real time on the card and in the AI Activity Dashboard:

QueuedThe mission is waiting in line to be processed by the agent.
RunningThe agent is actively executing the mission, streaming progress in real time via SSE.
Awaiting ApprovalThe agent has produced proposals and is waiting for a human to review them.
CompletedAll tasks have been executed successfully. Results are available in the mission log.
FailedThe mission encountered an error. The failure reason is logged for debugging.
CancelledA user manually cancelled the mission before it completed.

Proposals

When an agent runs in Supervised or Approval mode, it does not execute actions directly. Instead, it creates proposals — structured descriptions of the changes it wants to make.

Each proposal includes:

  • Action type — what the agent wants to do (create card, update field, post comment…)
  • Target — the entity that would be affected
  • Payload — the exact data the agent intends to write
  • Reasoning — the agent's explanation for why this change is needed

You can approve or reject each proposal individually. Approved proposals are executed immediately; rejected ones are discarded with an optional feedback note.

MCP Integration

Stellary exposes its agent tools via the Model Context Protocol (MCP), allowing external AI clients to connect directly to your workspace. Supported clients include:

  • Cursor — use Stellary tools from your IDE
  • Claude Code — connect your CLI-based AI assistant to your project board
  • Claude Desktop — interact with your workspace from the desktop app

The connection uses Streamable HTTP transport, so external clients receive real-time updates just like the built-in UI. Authentication is handled through workspace-scoped API tokens with the same permission model as regular agents.

Rate Limiting & Policies

Every agent operates within guardrails to prevent runaway behavior. Rate limits and policies work together to keep your workspace safe:

  • Rate limits — set the maximum number of tool calls an agent can make within a configurable time window. When the limit is reached, the agent pauses until the window resets.
  • Policy checks — every tool call is validated against workspace policies before execution. If a call violates a policy, it is blocked and logged.
  • Execution logging — every action, whether allowed or blocked, is recorded with full context: agent identity, tool name, parameters, result, and timestamp.

AI Activity Dashboard

The AI Activity Dashboard gives you a centralized view of everything your agents are doing. Accessible from the workspace sidebar, it provides:

  • Mission log — a chronological list of all missions with their current status, assigned agent, and target card
  • Tool call history — a detailed audit trail of every tool call, including parameters, results, and policy decisions
  • Agent stats — per-agent metrics showing mission count, success rate, and tool usage distribution
  • Pending proposals — a queue of proposals awaiting human review, grouped by agent and mission

Use the dashboard to monitor agent performance, identify issues early, and maintain full visibility over automated actions in your workspace.

User GuideBoard & CardsTeam & Collaboration
On this page
  • What are AI Agents?
  • Creating an Agent
  • Autonomy Modes
  • Agent Tools
  • Missions
  • Mission Lifecycle
  • Proposals
  • MCP Integration
  • Rate Limiting & Policies
  • AI Activity Dashboard