Any MCP host that supports the MCP authorization flow over Streamable HTTP. The Starko docs name Claude, Cursor and VS Code, and Claude Code connects with one command.

Can I use an API key instead of signing in?

No. The server only accepts tokens issued through Starko's consent screen, so every connection belongs to a person and a workspace and ends when that person leaves.

Can an MCP connection do more than I can in Starko?

No. Every call runs as you with your current role, through the same permission checks as the dashboard. Actions that need approval in Starko still need it over MCP.

Who can turn MCP access on?

Only the workspace owner. It is off by default. Switching it off cuts off every connected app at once.

What is not there yet?

There is no list of connected apps in Starko to revoke from yet; access ends from the host, by leaving the workspace, or by the owner's switch. MCP resources are not offered, and approvals use confirmation tokens rather than the protocol's elicitation.

Your Starko workspace,inside Claude, Cursor and VS Code.

The Starko MCP server gives any MCP host the tools the ⌘L chat uses: tickets, conversations, CRM, the knowledge base, reporting, workflows and Starko AI runs. Each person signs in with their own Starko account and acts with their own role, and every action that needs approval still asks for it.

ConnectorTypeStatus

Slack

Messaging

Connected

GitHub

northwind

Dev tools

Connected

Starko MCP server

Dev tools

Connected

Let Claude, Cursor, VS Code and other MCP apps use this workspace's Starko tools: the same tools as the ⌘L chat, with the same approvals. Each person connects with their own Starko login (OAuth) and acts with their own role.

Endpoint

https://v2.starko.one/api/mcp

Add to Claude Code

claude mcp add --transport http starko https://v2.starko.one/api/mcp

In Claude.ai, Cursor or VS Code, add a custom connector with the endpoint URL. The app opens Starko to sign in and pick the workspace; actions that need approval ask for it in the app.

Disable MCP access

v2.starko.one/dashboard

Settings/Integrations

72%

Integrations

Connect your workspace to external services: channels, scheduling and dev tools.

Search integrations

AllConnectedNot connected

ConnectorTypeStatus

Slack

Messaging

Connected

GitHub

northwind

Dev tools

Connected

Starko MCP server

Dev tools

Connected

Endpoint

https://v2.starko.one/api/mcp

Add to Claude Code

claude mcp add --transport http starko https://v2.starko.one/api/mcp

In Claude.ai, Cursor or VS Code, add a custom connector with the endpoint URL. The app opens Starko to sign in and pick the workspace; actions that need approval ask for it in the app.

Disable MCP access

At a glance

MCP server in short.

Endpoint: https://v2.starko.one/api/mcp
Transport: Streamable HTTP.
Sign-in: OAuth 2.1 with PKCE. There are no API keys.
Switch: Off by default. Only the workspace owner turns it on or off.
Who connects: Any member, with their own Starko login and their own role.
Scopes: tools:read and tools:write. Write includes read.
Approvals: A one-time confirmation token, valid for ten minutes.
Rate limit: 120 calls a minute per person and workspace.

Setup

Three steps, and each person signs in as themselves.

A workspace owner switches on MCP access in Settings, Integrations, under Starko MCP server. Then anyone on the team adds the endpoint to their host as a custom connector and clicks Connect. Starko opens its own consent screen: you sign in, pick the one workspace this connection may work in, see the role you will act as, and allow access.

In Claude Code it is one line: claude mcp add --transport http starko https://v2.starko.one/api/mcp.
In Claude.ai, Cursor or VS Code, add a custom connector with the endpoint URL.
The consent screen names the app, marks it Verified when it identifies itself with a metadata document, and says where your browser returns afterwards.
A connection works in one workspace. To use another, connect again and pick it.
The public docs list every tool and its access: v2.starko.one/docs/mcp.

Claude wants to connect to Starko

Choose the workspace it may work in. It will act as you, with your role, and every action that needs a person's approval will still ask for it.

Claude

claude.ai

Verified

WorkspaceOne per connection

NorthwindActing as Admin

Northwind LabsActing as Member

Access

Read your workspace

Tickets, conversations, CRM records, articles and reports you can see.

Act in your workspace

Create and change records as you. Replies, CRM changes and Starko AI runs still pause for approval.

After you decide, your browser returns to claude.ai.

CancelAllow access

Disconnecting the app on its side ends this access; it also ends when you leave the workspace.

Tools

The same tools as ⌘L, filtered by your role and scope.

The MCP server reads from the same tool registry as the ⌘L chat and Starko in Slack, so a tool behaves the same everywhere. Your host sees the tools your role allows, in a workspace configured for them. Read-only tools need tools:read; everything else needs tools:write.

Tickets and projects: search, read, create, update, comment, link, and create projects.
Customers: conversations, clients, companies, tenants and deals, replies and CRM changes.
Knowledge and reporting: articles, reporting, analytics, team workload and SLA status.
Factory: ticket workflows, recommendations, and Starko AI runs and sessions, including steering a session and merging its pull request.
Tools that only make sense in a chat are left out: navigating, asking you a question, follow-ups, chat attachments, exports and PDF reports.
Every tool is annotated read-only or destructive, so hosts that honour annotations can ask before calling.

Approvals

Writes with consequences never run on the first call.

A gated tool answers its first call with approval_required, a plain summary of what would happen and a one-time confirmation token. Your host shows the summary. When you agree, it repeats the call with the same arguments and the token, and the tool runs once. Change an argument and the token no longer matches.

Gated: replies and new e-mails to customers, CRM changes, conversation actions, article writes, archiving and batch ticket changes, creating, changing and running workflows, and starting or ending Starko AI runs, sessions and crews.
A token is valid for ten minutes and works once.
A tool that refuses answers ok: false with the reason, and the host is told to relay it as is.

1. tools/call assignStarkoAi

{ "ticketId": "8c1f…", "kind": "bug", "size": "s" }

Answer

{
  "ok": false,
  "code": "approval_required",
  "summary": "Start a Starko AI run on this ticket · bug, small",
  "confirmation_token": "c3RhcmtvLm…",
  "expires_in": 600
}

2. After the person agrees, same arguments

{ "ticketId": "8c1f…", "kind": "bug", "size": "s",
  "_confirm": "c3RhcmtvLm…" }

Answer

{ "ok": true, "data": { "status": "started", … } }

In Starko, a moment later

urgentIn ProgressBug Starko AI

CSV export times out for large accounts

Northwind Web

MNMia Novak2h

Skills

Your team's skills show up in the host's / menu.

Skills published in the Knowledge hub with MCP switched on appear as prompts in your host. Pick one and the host inserts the skill's instructions, plus your request if you add one. It is the same skill Starko uses in ⌘L, in Slack and in coding sessions.

Only published versions are served, never a draft.
Skills set to "Only Starko" stay out of the menu.
A skill's bundled files are listed by name.
Reconnect the connector or start a new conversation to see newly published skills.

Available in

⌘L chat

Slack

MCP

Coding sessions

MCP: shows up as a prompt in Claude, Cursor and VS Code (not for "Only Starko" skills).

Who can use it

People and StarkoIn the / menu, and Starko picks it when it fits

Only people, via /For procedures with side effects

Only StarkoBackground knowledge, hidden from the / menu

Access

Tokens tied to one person, one workspace and one app.

Starko is its own OAuth authorization server. Hosts register with a Client ID Metadata Document or dynamic registration, as public clients using PKCE. A token maps to one person and one workspace, and that person's role is read again on every call, so a change of role applies at once.

Access tokens last one hour. Refresh tokens last 30 days and rotate on every use.
Access ends when the host disconnects, when the person leaves the workspace, or when the owner switches MCP access off.
Each person gets 120 calls a minute per workspace. Past that, a call returns a short error asking to wait a minute.
Background and design notes are in the post Starko MCP server.

FAQAbout MCP server.

Connect a channel and a repository. Turn the support queue into a build queue.

Feedback starts flowing in minutes. The first recommendation shows up as soon as a pattern repeats, and the first pull request as soon as you accept one.

Start for free Get a demo

Your Starko workspace,inside Claude, Cursor and VS Code.

MCP server in short.

Three steps, and each person signs in as themselves.

The same tools as ⌘L, filtered by your role and scope.

Writes with consequences never run on the first call.

Your team's skills show up in the host's / menu.

Tokens tied to one person, one workspace and one app.

FAQAbout MCP server.

Which hosts work?

Can I use an API key instead of signing in?

Can an MCP connection do more than I can in Starko?

Who can turn MCP access on?

What is not there yet?

Keep exploring

Connect a channel and a repository. Turn the support queue into a build queue.