@dobbyai/mcp-external
Advanced tools
MCP Server for Dobby AI Platform - Connect from Claude Desktop, Cursor, Windsurf, and more
Connect to Dobby AI Platform from Claude Desktop, Cursor, Windsurf, VS Code, and any MCP-compatible client.
npx @dobbyai/mcp-external --setup
This will guide you through connecting Dobby to your preferred AI tools.
{
"mcpServers": {
"dobby": {
"command": "npx",
"args": ["-y", "@dobbyai/mcp-external"],
"env": {
"DOBBY_API_URL": "https://dobby-ai.com",
"DOBBY_API_KEY": "gk_user_your_key_here",
"DOBBY_TENANT_ID": "your-tenant-id"
}
}
}
}
| Client | Status | Setup Command |
|---|---|---|
| Claude Desktop | ✅ Full Support | npx @dobbyai/mcp-external --setup |
| Cursor | ✅ Full Support | npx @dobbyai/mcp-external --setup |
| Windsurf | ✅ Full Support | npx @dobbyai/mcp-external --setup |
| VS Code (Cline) | ✅ Full Support | Manual config |
| VS Code (Continue) | ✅ Full Support | Manual config |
| Any MCP Client | ✅ STDIO Mode | Manual config |
| Tool | Description | Example |
|---|---|---|
create_task | Create a new task for AI agents | create_task(title: "Add dark mode", priority: "high") |
list_tasks | List tasks with filters | list_tasks(status: "pending", limit: 10) |
get_task | Get detailed task information | get_task(task_id: "task-abc123") |
update_task | Update task status or priority | update_task(task_id: "...", status: "in_progress") |
get_task_timeline | Get activity history for a task | get_task_timeline(task_id: "...") |
cancel_task | Cancel a running task | cancel_task(task_id: "...") |
| Tool | Description | Example |
|---|---|---|
get_messages | Get conversation for a task | get_messages(task_id: "...", limit: 20) |
send_message | Send a message to task conversation | send_message(task_id: "...", content: "Please add tests") |
| Tool | Description | Example |
|---|---|---|
get_pending_approvals | List approvals waiting for you | get_pending_approvals() |
approve_task | Approve a pending request (auto-merges PR) | approve_task(approval_id: "...", note: "LGTM") |
reject_task | Reject with feedback (agent retries automatically) | reject_task(approval_id: "...", reason: "Need tests") |
request_changes | Request specific modifications | request_changes(approval_id: "...", changes: "...") |
| Tool | Description | Example |
|---|---|---|
list_services | List all available services and capabilities | list_services(category: "llm") |
| Tool | Description | Example |
|---|---|---|
crewai_health | Check health of connected CrewAI agent | crewai_health() |
crewai_inputs | Get required input fields for a crew | crewai_inputs() |
crewai_kickoff | Start a CrewAI crew execution | crewai_kickoff(inputs: { topic: "AI agents" }) |
crewai_status | Check status of a crew execution | crewai_status(kickoff_id: "...") |
| Tool | Description | Example |
|---|---|---|
crew_validate | Validate crew YAML configurations | crew_validate(agents_yaml: "...", tasks_yaml: "...") |
crew_kickoff | Create and start a crew from YAML | crew_kickoff(agents_yaml: "...", tasks_yaml: "...") |
crew_status | Check status of a local crew run | crew_status(run_id: "...") |
crew_list | List recent crew runs | crew_list(limit: 10) |
| Tool | Description | Example |
|---|---|---|
mcp_discover_tools | Discover tools on a remote MCP server | mcp_discover_tools(provider_id: "langsmith") |
mcp_call_tool | Call a tool on a remote MCP server | mcp_call_tool(tool_name: "...", tool_arguments: {...}) |
| Tool | Description | Example |
|---|---|---|
jira_search_issues | Search issues using JQL | jira_search_issues(jql: "project = DEV AND status = Open") |
jira_get_issue | Get detailed issue information | jira_get_issue(issue_key: "DEV-123") |
jira_create_issue | Create a new Jira issue | jira_create_issue(project_key: "DEV", summary: "...") |
jira_update_issue | Update an existing issue | jira_update_issue(issue_key: "DEV-123", summary: "...") |
jira_add_comment | Add a comment to an issue | jira_add_comment(issue_key: "DEV-123", body: "...") |
jira_get_comments | Get comments on an issue | jira_get_comments(issue_key: "DEV-123") |
jira_get_transitions | Get available workflow transitions | jira_get_transitions(issue_key: "DEV-123") |
jira_transition_issue | Transition issue to a new status | jira_transition_issue(issue_key: "DEV-123", transition_id: "31") |
jira_list_projects | List all accessible Jira projects | jira_list_projects() |
jira_assign_issue | Assign an issue to a user | jira_assign_issue(issue_key: "DEV-123", account_id: "...") |
The Dobby MCP Server supports a complete Human-in-the-Loop workflow:
┌─────────────────────────────────────────────────────────────────┐
│ APPROVAL WORKFLOW │
├─────────────────────────────────────────────────────────────────┤
│ │
│ 1. Agent completes work → Creates PR → Requests approval │
│ │
│ 2. User receives notification via get_pending_approvals │
│ │
│ 3. User reviews and decides: │
│ │
│ ┌─────────────────┐ ┌─────────────────┐ │
│ │ approve_task │ │ reject_task │ │
│ │ │ │ │ │
│ │ ✅ Auto-merge │ │ 🔁 Auto-retry │ │
│ │ PR │ │ with feedback │ │
│ └────────┬────────┘ └────────┬────────┘ │
│ │ │ │
│ ▼ ▼ │
│ ┌─────────────────┐ ┌─────────────────┐ │
│ │ Task Completed │ │ Agent Addresses │ │
│ │ 🎉 │ │ Feedback & ... │──► Back to 1 │
│ └─────────────────┘ └─────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────────┘
approve_task(approval_id, note?)completedget_task_timelinereject_task(approval_id, reason)Track every action on a task with get_task_timeline:
📊 Timeline for Task task-abc123
🆕 task created (User)
2/2/2026, 10:00:00 AM
📋 plan created (DevCoordinator)
2/2/2026, 10:00:05 AM
▶️ subtask started (dobby-frontend-agent)
2/2/2026, 10:00:10 AM
🎨 frontend code generated (FrontendDeveloper)
2/2/2026, 10:01:30 AM
📝 pr created (FrontendDeveloper)
PR #42
2/2/2026, 10:01:45 AM
⏳ approval requested (DevCoordinator)
2/2/2026, 10:01:50 AM
✅ approval approved (User)
2/2/2026, 10:05:00 AM
🚀 agent continuation triggered (system)
2/2/2026, 10:05:01 AM
🔀 pr merged (DevCoordinator)
PR #42
2/2/2026, 10:05:10 AM
🎉 continuation after approval completed (DevCoordinator)
2/2/2026, 10:05:15 AM
Access data directly via MCP resources:
tasks://list # All tasks
tasks://pending # Pending tasks
tasks://in-progress # In-progress tasks
tasks://completed # Completed tasks
tasks://{task_id} # Specific task details
tasks://{task_id}/messages # Task conversation
approvals://pending # Pending approvals
| Command | Description |
|---|---|
/create-feature | Guided feature creation |
/fix-bug | Create bug fix task |
/check-approvals | Review pending approvals |
/task-status | Get task status summary |
/review-pr | Review a task's PR |
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json
{
"mcpServers": {
"dobby": {
"command": "npx",
"args": ["-y", "@dobbyai/mcp-external"],
"env": {
"DOBBY_API_URL": "https://dobby-ai.com",
"DOBBY_API_KEY": "gk_user_your_key_here",
"DOBBY_TENANT_ID": "your-tenant-id"
}
}
}
}
Settings → MCP → Add Server:
{
"dobby": {
"command": "npx",
"args": ["-y", "@dobbyai/mcp-external"],
"env": {
"DOBBY_API_URL": "https://dobby-ai.com",
"DOBBY_API_KEY": "gk_user_your_key_here",
"DOBBY_TENANT_ID": "your-tenant-id"
}
}
}
File: ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"dobby": {
"command": "npx",
"args": ["-y", "@dobbyai/mcp-external"],
"env": {
"DOBBY_API_URL": "https://dobby-ai.com",
"DOBBY_API_KEY": "gk_user_your_key_here",
"DOBBY_TENANT_ID": "your-tenant-id"
}
}
}
}
Command Palette → "Cline: Open MCP Settings" → Add:
{
"mcpServers": {
"dobby": {
"command": "npx",
"args": ["-y", "@dobbyai/mcp-external"],
"env": {
"DOBBY_API_URL": "https://dobby-ai.com",
"DOBBY_API_KEY": "gk_user_your_key_here",
"DOBBY_TENANT_ID": "your-tenant-id"
}
}
}
}
| Variable | Required | Description |
|---|---|---|
DOBBY_API_URL | Yes | Dobby API URL (default: https://dobby-ai.com) |
DOBBY_API_KEY | Yes | Your Dobby key: a gk_user_/gk_svc_ gateway key, an sk_live_ tenant key, or (remote HTTP transport only) a dobby_pat_ Personal Access Token. See Credentials & Transports below. |
DOBBY_TENANT_ID | Yes | Your workspace/tenant ID |
DOBBY_ORGANIZATION_ID | No | Organization ID for org-scoped tools (memory stores, rubrics, orchestration configs). The variable name is DOBBY_ORGANIZATION_ID — a DOBBY_ORG_ID variable is not read and is silently ignored. |
DOBBY_USER_ID | No | User ID for messages |
DOBBY_USER_NAME | No | Display name for messages |
DOBBY_GATEWAY_KEY | No | Optional — auto-detected if DOBBY_API_KEY starts with gk_; set explicitly to override. |
DOBBY_GATEWAY_URL | No | Optional — defaults to DOBBY_API_URL. |
DOBBY_REGION | No | Optional — data-residency region hint. |
Not sure what your key can do or which org/tenant you're on? Run
npx @dobbyai/mcp-external --doctorbefore connecting, or — once connected — call thewhoamitool. Both report your transport, key type, resolved org/tenant, whether a PAT is present, and exactly what (if anything) is missing.
Dobby MCP runs in two transports, and which credential you need depends on which one you use. Run npx @dobbyai/mcp-external --doctor at any time to check your setup — it prints exactly what your credentials can and cannot do.
The default (npx @dobbyai/mcp-external), used by Claude Desktop, Cursor, Windsurf, etc. Configured with env vars:
DOBBY_API_KEY value | Read / tenant tools | Gateway / execute tools |
|---|---|---|
gk_user_… / gk_svc_… (gateway key) | ✅ | ✅ — auto-enabled; a gk_ key doubles as the gateway key |
sk_live_… (tenant key) | ✅ | ❌ — also set a gk_ key in DOBBY_GATEWAY_KEY |
dobby_pat_… (Personal Access Token) | ✅ | ❌ — not available in STDIO (see below) |
Gateway/execute tools in STDIO need a gk_ key — either as DOBBY_API_KEY or as DOBBY_GATEWAY_KEY. A Personal Access Token (dobby_pat_) cannot drive gateway tools locally: minting the short-lived gateway key it would need requires an internal secret that only exists on Dobby-operated servers, never on a customer machine — by design.
Point an MCP client at https://mcp.dobby-ai.com and authenticate with OAuth or a Personal Access Token (dobby_pat_). Here a single PAT covers everything — read, tenant, org, and gateway tools — because the server mints a per-request, tenant-scoped gk_tmp_ from your session on your behalf. No separate gateway key to manage.
Rule of thumb: local STDIO → use a
gk_user_key (works for both read and execute). Remote → use a PAT (one credential, gateway included).
You: Create a task to add dark mode to the settings page
Claude: [Uses create_task tool]
✅ Task created: task-abc123
Title: Add dark mode to settings page
Priority: medium
The AI agents will start working on it soon.
You: Check my pending approvals
Claude: [Uses get_pending_approvals]
🔔 You have 2 pending approvals:
1. 📝 PR CREATE - Add user authentication
ID: approval-xyz789
Waiting: 5 minutes
2. 🔀 PR MERGE - Fix login bug
ID: approval-abc456
Waiting: 1 hour
Which one would you like to review?
You: Approve approval-xyz789 with note "looks good"
Claude: [Uses approve_task]
✅ **Approved!** approval-xyz789
Note: "looks good"
🔀 The agent is now merging the PR automatically.
Use `get_task_timeline` to track progress.
You: Reject approval-abc456 because it needs unit tests
Claude: [Uses reject_task]
❌ **Rejected!** approval-abc456
💬 Feedback: "needs unit tests"
🔁 The agent is now retrying and will address your feedback.
Use `get_task_timeline` to track progress.
You: Cancel task-def789, it's taking too long
Claude: [Uses cancel_task]
🛑 Task task-def789 execution cancelled successfully.
The task status has been reverted to pending.
You: Show me the timeline for task-abc123
Claude: [Uses get_task_timeline]
📊 Timeline for Task task-abc123
🆕 **task created** (User)
2/2/2026, 10:00:00 AM
📋 **plan created** (DevCoordinator)
2/2/2026, 10:00:05 AM
🎨 **frontend code generated** (FrontendDeveloper)
2/2/2026, 10:01:30 AM
📝 **pr created** PR #42
2/2/2026, 10:01:45 AM
✅ **approval approved**
2/2/2026, 10:05:00 AM
🔀 **pr merged** PR #42
2/2/2026, 10:05:10 AM
┌─────────────────────────────────────────────────────────────────┐
│ MCP Client │
│ (Claude Desktop, Cursor, Windsurf, etc.) │
└──────────────────────────┬──────────────────────────────────────┘
│ MCP Protocol (STDIO)
▼
┌─────────────────────────────────────────────────────────────────┐
│ Dobby MCP Server │
│ ┌─────────────────────────────────────────────────────────────┐│
│ │ 36 Tools │ Resources │ Prompts ││
│ │ │ │ ││
│ │ Tasks (6) │ - tasks:// │ - /create-feature ││
│ │ Messages (2) │ - approvals:// │ - /fix-bug ││
│ │ Approvals (4) │ - services:// │ - /check-approvals ││
│ │ Services (1) │ │ - /task-status ││
│ │ CrewAI (4) │ │ - /review-pr ││
│ │ Crew Builder (4)│ │ ││
│ │ MCP Remote (2) │ │ ││
│ │ Jira Cloud (10) │ │ ││
│ │ + 3 more │ │ ││
│ └─────────────────────────────────────────────────────────────┘│
│ │ │
│ DobbyClient (API) │
└──────────────────────────────┬──────────────────────────────────┘
│ HTTPS
▼
┌─────────────────────────────────────────────────────────────────┐
│ Dobby API (Next.js) │
│ /api/dobby/tasks - Task CRUD │
│ /api/dobby/tasks/[id]/execute - Execute/Cancel task │
│ /api/dobby/tasks/[id]/conversation - Messages │
│ /api/dobby/approvals - Approval workflow │
└──────────────────────────────┬──────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ Dobby Python API (FastAPI) │
│ POST /api/v1/tasks/{id}/execute - Start agent execution │
│ POST /api/v1/tasks/{id}/cancel - Cancel execution │
│ GET /api/v1/runs/{id}/status - Get run status │
└──────────────────────────────┬──────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ AI Agents (Dobby) │
│ ┌──────────────┐ │
│ │ Chief Agent │ ─► Analyzes tasks, routes to coordinators │
│ └──────┬───────┘ │
│ │ │
│ ┌──────▼───────┐ │
│ │ DevCoordinator│ ─► Manages development workflow │
│ └──────┬───────┘ │
│ │ │
│ ┌──────▼───────────────────────────────────────┐ │
│ │ Workers │ │
│ │ - FrontendDeveloper (React, TypeScript) │ │
│ │ - BackendDeveloper (Python, FastAPI) │ │
│ │ - TestWriter (Unit tests) │ │
│ │ - CodeReviewer (PR review) │ │
│ │ - DevOpsWorker (CI/CD, PR merge) │ │
│ └───────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────┘
# Clone the repo
git clone https://github.com/gil-dataopsvalley/repo-dobby.git
cd repo-dobby/mcp/external
# Install dependencies
npm install
# Run in development mode
npm run dev
# Build
npm run build
# Run tests
npm test
# Set environment variables
export DOBBY_API_URL=http://localhost:3000
export DOBBY_API_KEY=gk_user_your_key_here # local dev server only (auth bypassed)
export DOBBY_TENANT_ID=your-tenant-id
# Run the server
npm run dev
get_task_timeline tool - View all activity on a taskcancel_task tool - Stop running task executionsapprove_task now triggers automatic PR mergereject_task now triggers automatic agent retry with feedbackMIT © Dobby AI Platform
FAQs
MCP Server for Dobby AI Platform - Connect from Claude Desktop, Cursor, Windsurf, and more
We found that @dobbyai/mcp-external demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Research
/Security News
Socket uncovered two malicious VS Code themes in a GlassWorm-linked cluster with thousands of installs across VS Code Marketplace and Open VSX.
Security News
/Company News
Capital One is partnering with Socket to proactively secure its open source supply chain.
Security News
Socket CTO Ahmad Nassri discusses how to keep AI agents from bypassing package blocks, limit credential access, and monitor their actions.