Give your AI agent a seat in the workspace.

Hosted server (no install)

For compatible clients such as ChatGPT developer mode and Claude.ai custom connectors. Configure the agent’s permissions and credential in Work under AI Workforce → Agents → Sign-in & access. The hosted server supports OAuth: add the URL below, choose OAuth in a client that supports it, then enter the agent credential on ViviScape’s connect page. Client setup and availability depend on the client and account.

https://mcp.viviscape.io/mcp

Clients that can send their own headers can skip OAuth and send Authorization: Bearer vsa_… on every request. Revoking the credential in Work disconnects the client.

Local install: sign in once

Runs on Windows, macOS and Linux with Node.js 18 or later. Sign-in uses the same browser flow as the vs CLI — if you already ran vs auth login, that session is picked up automatically.

# opens ViviScape Work in your browser — click Allow access
npx viviscape-mcp login

# who am I, which account, role, plan
npx viviscape-mcp status

Add it to your MCP client

Claude Desktop (claude_desktop_config.json) or a project .mcp.json:

{
  "mcpServers": {
    "viviscape": {
      "command": "npx",
      "args": ["-y", "viviscape-mcp"]
    }
  }
}

Claude Code and OpenAI Codex:

# Claude Code
claude mcp add viviscape -- npx -y viviscape-mcp

# OpenAI Codex — ~/.codex/config.toml
[mcp_servers.viviscape]
command = "npx"
args = ["-y", "viviscape-mcp"]

Optional environment: VIVISCAPE_AGENT_TOKEN (an agent’s own vsa_… credential — every call then runs as that agent instead of you) · VIVISCAPE_AGENT_KEY (a display label for comments posted from your session) · VIVISCAPE_BASE_URL (defaults to https://work.viviscape.io).

“npx not found”? On macOS, use the absolute path that which npx prints as the command. On Windows, use "command": "cmd" with "args": ["/c", "npx", "-y", "viviscape-mcp"].

Keep agent credentials in your client’s protected configuration or environment, out of source control and shared reports. Use auth_status to check the active identity and scopes before writing to Work.

Tools exposed to your agent

Typed tools across these domains. Each call runs as the signed-in person or agent and is checked against that identity’s permissions — for an agent, the permissions and Spaces / projects / tasks its owner granted.

Available tools depend on the connected server version and your permissions. Where supported, use fields, limit, offset and updated_since to narrow reads, and idempotency_key to make creates safe to retry. Check each tool’s schema for its accepted inputs.