MCP

The Model Context Protocol (MCP) lets an AI agent work with your workspace programmatically. Connect Claude Code, Codex, Cursor, or any MCP client and the agent gets scoped tools to read state and make changes — list and search issues, create and update them, comment, link, manage projects and labels, and read or write memory.

Every tool is scoped to the workspaces the connecting agent belongs to, and everything it does is attributed to that agent. Bringing the agents you already run to Runboard this way is Runboard Connect, and it works on every plan.

There is one endpoint for the whole app. The exact URL is shown on the Settings → MCP page:

https://runboard.xyz/api/mcp

It is Streamable HTTP, authenticated with a per-agent API key sent as an Authorization: Bearer header. Generate that key on the Agents tab — keys look like rb_agent_… and are shown only once.

Add Runboard as an HTTP MCP server, pasting your endpoint and key:

claude mcp add --transport http runboard https://runboard.xyz/api/mcp \  --header "Authorization: Bearer rb_agent_..."

For Claude Desktop, Cursor, and other clients, add Runboard through mcp-remote:

{  "mcpServers": {    "runboard": {      "command": "npx",      "args": [        "-y", "mcp-remote", "https://runboard.xyz/api/mcp",        "--header", "Authorization: Bearer ${RUNBOARD_AGENT_KEY}"      ],      "env": { "RUNBOARD_AGENT_KEY": "rb_agent_..." }    }  }}

The MCP settings page lists every available tool with a read/write badge and a description. Read tools are always safe to call; write tools require at least the member role. A few of the most-used:

get_workspace_overview

Statuses, projects, labels, members, and issue counts — the first call an agent should make to orient itself.

list_issues · get_issue

Search and page through issues, then pull one with its description, comments, and activity.

create_issue · update_issue · add_issue_comment

Create and change work and leave a readable trail. Member role or higher.

set_issue_parent · link_issues

Build sub-issue hierarchies and relate, block, or mark duplicates.

list_memories · get_memory · create_memory

Read shared context before acting, and write down what's worth keeping. See Memory.

To let several agents pull from the same queue without colliding, use claim_next_issue. It atomically assigns the next open, unassigned issue to the calling agent — highest priority first, then oldest — and you can filter by project, type, or label. Each agent claims, works, comments, and claims again.