AgentBuilder MCP Server

Package: @chandshantanu/agentbuilder-mcp-server Version: 0.2.0 License: MIT

A Model Context Protocol (MCP) server that lets Claude Code create and manage AI agents, workflows, and dashboards in the AgentBuilder / Chatslytics platform — using natural language.

Quick Setup

No source code needed. Just Node.js ≥ 18.

1. Add to Claude Code

claude mcp add agentbuilder -- npx -y @chandshantanu/[email protected]

2. Restart Claude Code

exit   # or Ctrl+C
claude

3. First Run — Browser Login

The first time you use an AgentBuilder tool, the server will prompt:

AgentBuilder login required.
Opening browser at: https://www.chatslytics.com/login?mcp_callback=http://127.0.0.1:<PORT>/callback
Waiting for login (5 minute timeout)...

Log in with your Chatslytics account. The browser completes authentication automatically — no token copy-pasting needed.

Credentials are cached at ~/.agentbuilder/credentials.json and reused in future sessions.

How It Works

• Runs over stdio (standard MCP transport) — no HTTP server needed
• Uses npx — the package is downloaded and cached locally on first use, no global install required
• Browser-based login: opens www.chatslytics.com/login with a local callback server to capture the Supabase JWT automatically
• Token is stored encrypted and reused until expiry

Available Tools

Agent Management

| Tool | Description | |------|-------------| | create_agent | Create a new agent with metadata | | get_agent | Get agent details | | list_agents | List agents with optional filters | | update_agent | Update agent metadata | | publish_agent | Publish agent to marketplace | | deploy_agent | Deploy agent to Kubernetes pod | | redeploy_agent | Redeploy existing agent | | validate_agent_configuration | Validate before publish | | set_agent_wizard_config | Configure onboarding wizard (OAuth, credentials, fields) | | set_agent_graph | Set LangGraph workflow definition (nodes + edges) |

Workflow Management

| Tool | Description | |------|-------------| | create_workflow | Create a LangGraph workflow | | list_workflows | List all workflows | | get_workflow | Get workflow details | | update_workflow | Modify a workflow | | delete_workflow | Remove a workflow | | execute_workflow | Run a workflow with input | | validate_workflow | Check structure before saving | | get_execution_status | Check execution result | | list_executions | List execution history |

Agent Composition

| Tool | Description | |------|-------------| | compose_agents | Link multiple agents together | | get_composed_agent | Get composed agent details | | list_composed_agents | List all composed agents | | preview_composition | Preview before committing | | execute_composed_agent | Run a composed agent pipeline |

Dashboard Configuration

| Tool | Description | |------|-------------| | create_agent_dashboard | Create a dashboard for an agent | | update_agent_dashboard | Update dashboard config | | get_agent_dashboard_config | Get dashboard config | | configure_agent_page | Configure landing/config/dashboard pages | | add_dashboard_widget | Add a widget to a dashboard | | generate_dashboard_widget | AI-generate a custom widget (TSX) | | list_generated_widgets | List generated widgets | | regenerate_widget | Regenerate with new prompt | | delete_generated_widget | Delete a generated widget |

Platform

| Tool | Description | |------|-------------| | set_auth_token | Manually set a Supabase JWT (alternative to browser login) | | set_connection | Switch workspace/connection | | get_mcp_plan_limits | Check platform tier limits | | list_mcp_workflows | List MCP-compatible workflows |

Manual Token (Alternative)

If you prefer to set a token manually instead of the browser flow:

1. Log in at www.chatslytics.com
2. Open DevTools → Application → Local Storage → find supabase.auth.token
3. Copy the access_token value
4. In Claude Code: Set my agentbuilder auth token: eyJ...

Or use the set_auth_token MCP tool directly.

Example Usage

Once connected, use natural language in Claude Code:

Create an Instagram DM agent that handles product inquiries

Show me all my agents

Deploy agent <agent_id> for subscription <subscription_id>

Create a workflow that:
- Triggers on webhook
- Classifies message intent
- Generates an LLM response
- Sends reply via WhatsApp

Troubleshooting

"Login prompt appeared but browser didn't open" → Manually navigate to the URL printed in the terminal.

"Token expired" → Delete ~/.agentbuilder/credentials.json and re-run to trigger a fresh login.

"Tools not showing in Claude Code" → Run claude mcp list to confirm agentbuilder is registered. If missing, re-run the claude mcp add command.

"npx too slow" → After first run, npx uses the cached version. Subsequent starts are fast.

Requirements

• Node.js ≥ 18
• Claude Code CLI
• An account at www.chatslytics.com

Links

• Platform: www.chatslytics.com
• MCP Protocol: modelcontextprotocol.io
• npm package: npmjs.com/package/@chandshantanu/agentbuilder-mcp-server