What You Can Do

Manage tasks from your IDE

Create, assign, and track agent tasks without leaving Claude Desktop or Cursor.

Approve agent actions

Review and approve pending agent requests directly from your development environment.

Monitor agent health

Check service status, view active issues, and access the operations audit trail.

Automate workflows

Search GitHub/Jira, manage schedules, run CrewAI crews, and query knowledge bases.

Quick Setup

Prerequisites

You need a Dobby account with a Gateway API key. Keys are created in your workspace dashboard under Settings > Gateway.

1Open Claude Desktop settings

2Navigate to Developer > Edit Config

3Paste the configuration below

4Restart Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json | Windows: %APPDATA%\Claude\claude_desktop_config.json

JSON

{
  "mcpServers": {
    "dobby": {
      "command": "npx",
      "args": [
        "-y",
        "@dobbyai/mcp-external"
      ],
      "env": {
        "DOBBY_API_URL": "https://dobby-ai.com",
        "DOBBY_API_KEY": "<YOUR_GATEWAY_KEY>",
        "DOBBY_TENANT_ID": "<YOUR_TENANT_ID>"
      }
    }
  }
}

Replace <YOUR_GATEWAY_KEY>, <YOUR_TENANT_ID>, and <YOUR_ORG_ID> with your actual credentials from the dashboard.

Available Tools

149 tools across 33 categories. Click a category to see its tools.

Available by plan (cumulative): MVP 59 · Pro 107 · Enterprise 149

Tool Access by Plan

The remote server enforces the catalog above by your organization's subscription plan. The tier badge on each tool shows the plan it requires, tiers are cumulative, and the same rule applies to every authentication method — OAuth login, Gateway keys, and tenant API keys alike.

Plan	Tool Tiers	Tools You'll See
Free / Trial	MVP	61 core tools
Pro / Team	MVP + Pro	109 tools
Enterprise	All tiers	144 tools

The catalog holds 149 tools. Five are Dobby platform-operations tools reserved for our own staff, so the most any customer account lists is 144.

Read-only keys — a key whose scopes all end in :read gets only the read-only tools within your plan's tier. Use one for dashboards or agents that should never mutate state.

Out-of-tier calls — tools above your plan don't appear in the tool list, and calling one directly returns an error result naming the plan that unlocks it. Nothing executes.

Upgrading — move your organization to a higher plan and reconnect (or refresh your client's tool list): the additional tools appear on the same server URL with the same key, no reconfiguration.

Remote Connector (One-Click)

Connect to Dobby from Claude Code, Cursor, or VS Code without installing anything locally. Uses OAuth 2.1 — just enter the server URL and log in.

Server URL

https://mcp.dobby-ai.com/mcp

Use exactly that URL. Enter https://mcp.dobby-ai.com/mcp — not the address of this documentation page (dobby-ai.com/docs/mcp) and not the main site (dobby-ai.com). Those let you sign in but then fail with "authorized, but error connecting" — the MCP endpoint lives only on the mcp. subdomain.

How it works: Enter the URL above in your IDE's MCP connector settings. A browser window opens for OAuth login. Select your workspace, click "Allow", and your plan's tools are available instantly (see Tool Access by Plan above). No API keys needed.

Or connect with an API key

For headless, scripted, or CI setups where a browser login isn't practical, skip OAuth and pass a tenant API key (sk_live_*) or Gateway key (gk_*) as a static Bearer token:

claude mcp add --transport http dobby https://mcp.dobby-ai.com/mcp \
  --header "Authorization: Bearer sk_live_YOUR_KEY"

Or in any MCP client that supports HTTP headers:

{
  "mcpServers": {
    "dobby": {
      "type": "http",
      "url": "https://mcp.dobby-ai.com/mcp",
      "headers": { "Authorization": "Bearer sk_live_YOUR_KEY" }
    }
  }
}

The key authenticates every request directly — no browser, no token refresh. Create keys in your workspace dashboard under Settings.

For LLM Agents & Scripts (Raw HTTP)

An autonomous agent or script doesn't need an MCP client library — the server speaks Streamable HTTP JSON-RPC directly. Four rules:

Send Authorization: Bearer <sk_live_ or gk_ key> on every request.
Include Accept: application/json, text/event-stream — responses may arrive as an SSE frame (data: {json}).
Capture the Mcp-Session-Id response header from initialize and echo it on every later call — the session is stateful and bound to your key. Missing it → 404 Session not found.
Handshake order: initialize → notifications/initialized → then tools/list / tools/call.

# 1. initialize — grab Mcp-Session-Id from the response headers
curl -sS -D - -X POST https://mcp.dobby-ai.com/mcp \
  -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"my-agent","version":"1.0"}}}'

# 2. call a tool (echo the session id)
curl -sS -X POST https://mcp.dobby-ai.com/mcp \
  -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" -H "Mcp-Session-Id: $SID" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"list_cp_reports","arguments":{"limit":3}}}'

A result like "No Control Plane reports found" means the call succeeded — the workspace just has no data yet.

Authentication

All MCP requests are authenticated via your Gateway API key. Three key tiers are available:

Prefix	Type	Rate Limit	Best For
gk_user_*	User	100 RPM	IDE / personal use
gk_svc_*	Service	500 RPM	Automated agents / CI
gk_tmp_*	Temporary	50 RPM	Testing / demos (auto-expire)

MCP tools require mcp:read and mcp:execute scopes. These are included by default when creating a key.

The remote connector (https://mcp.dobby-ai.com/mcp) additionally accepts your tenant API key (sk_live_*) as a static Bearer token — the same key you use for the REST API.

Security

Authenticated access — every MCP request validates your Gateway key with scope-level permissions

Full audit trail — every tool invocation is logged with actor, timestamp, and result

Regional data residency — your data stays in your selected region (Israel, EU, or US)

Kill-switch — organization admins can instantly block all MCP access via the kill-switch

MCP Server