Connect an MCP client to live project data, claim agent missions, update cards, read documents, and keep every governed action inside the Stellary workspace.
AI client
Claude · Cursor · Codex
Stellary MCP
Streamable HTTP
Launch workspace
Cards · Docs · Missions
POST https://api.stellary.co/mcp
Authorization: Bearer ••••••••••••
200 · tools/list · workspace context resolved
Stellary exposes a remote Streamable HTTP endpoint. Identity, workspace permissions, project access, tool policy, and autonomy mode determine what each caller can do.
01
Use a personal access token for a human client or an agent token for a dedicated runner.
02
List projects, cards, documents, cockpit signals, or queued missions.
03
Reuse exact IDs instead of relying on ambiguous names.
04
Call the tools permitted for that identity and project scope.
05
Return completion, failure, proposal, or progress to the workspace.
Mission STL-128
Use a personal access token for a human client or an agent token for a dedicated runner.
Checklist
Linked context
A user JWT or personal access token is suited to interactive clients acting as the signed-in person.
The endpoint is not a documentation-only wrapper. It reaches the same project objects, mission queues, policy checks, and installed workspace tools used by Stellary agents.
A user JWT or personal access token is suited to interactive clients acting as the signed-in person.
An agent token carries workspace context, agent rules, tool access, and the configured autonomy mode.
External agents can initialize, wait for work, claim a mission, and report completion or failure.
{
"mcpServers": {
"stellary": {
"url": "https://api.stellary.co/mcp",
"headers": {
"Authorization": "Bearer YOUR_TOKEN"
}
}
}
}| Action | Human | Supervised | Approval |
|---|---|---|---|
| Interactive project exploration | Recommended | Possible | Possible |
| Queued agent missions | No | Recommended | Recommended |
| Workspace plugin tools | Limited | Allowlist | Proposal |
| Non-read writes | Direct | By policy | Proposal |
Use MCP when an AI client needs to discover context and choose tools during a reasoning loop. Use REST when an integration needs deterministic resource provisioning, explicit payloads, or a conventional backend-to-backend contract.
Both surfaces operate on the same project system. MCP does not create a second copy of the board or a shadow task database.
Authentication only establishes identity. Stellary still checks workspace membership, project access, agent tool allowlists, scope, and autonomy policy before a tool changes shared data.
In approval mode, non-read agent tools create persisted proposals. A human can inspect and approve the requested action from the workspace.
REST API
Stellary MCP
Caller
REST API
Deterministic service or script.
Stellary MCP
Interactive AI client or agent loop.
Discovery
REST API
Routes and payloads selected in code.
Stellary MCP
Tools discovered and selected at runtime.
Provisioning
REST API
Best for explicit CRUD and setup.
Stellary MCP
Best after the workspace context exists.
Agent policy
REST API
Application implements its own loop.
Stellary MCP
Agent identity and autonomy policy follow the call.
When enabled for the workspace, Stellary MCP uses Streamable HTTP at https://api.stellary.co/mcp. It accepts GET for discovery or negotiation and POST for MCP requests.
Any compatible client that supports a remote Streamable HTTP server and authorization headers can connect. The documentation includes examples for Cursor, Claude Code, and Claude Desktop-style configurations.
Use a personal access token when the client acts interactively as you. Use an agent token for a dedicated workspace agent that needs queued missions, workspace tool context, and agent-specific policy.
Yes, when its identity, project access, tool allowlist, and autonomy policy permit the action. Approval mode converts non-read agent actions into proposals instead of applying them directly.
Create a workspace, issue the right token, and start from one exact project.
Create a Stellary workspace