LivePerson MCP Server

MCP server that gives AI coding assistants direct access to LivePerson's Conversational Cloud APIs. Manage knowledge bases, bots, conversations, users, skills, and more — through natural language.

Works with any AI tool that supports the Model Context Protocol: Claude Code, OpenAI Codex, Gemini CLI, VS Code Copilot, Cursor, Windsurf, and others.

Prerequisites

• Node.js 18+
• A LivePerson account with a bot-type user and API keys

Setup

1. Create a bot user in LivePerson

1. Go to Conversational Cloud > Manage > Users
2. Create a new user (e.g. claude-mcp)
3. Set type to Bot
4. Assign the Admin profile (or a custom profile with the permissions you need)
5. Generate API keys for the user — you'll get:
   • App Key
   • Secret
   • Access Token
   • Access Token Secret

2. Install and configure

npx @lpextend/mcp-server --setup

This prompts for your credentials and saves them to ~/.config/lp-mcp/.env.

Or create the file manually:

LP_ACCOUNT_ID=12345678
LP_LOGIN_NAME=claude-mcp
LP_APP_KEY=your-app-key
LP_SECRET=your-secret
LP_ACCESS_TOKEN=your-access-token
LP_ACCESS_TOKEN_SECRET=your-access-token-secret

3. Connect to your AI tool

Claude Code

claude mcp add liveperson -- npx @lpextend/mcp-server

Or add to your project's .mcp.json:

{
  "mcpServers": {
    "liveperson": {
      "command": "npx",
      "args": ["@lpextend/mcp-server"]
    }
  }
}

OpenAI Codex

codex mcp add liveperson -- npx @lpextend/mcp-server

Or add to ~/.codex/config.toml:

[mcp_servers.liveperson]
command = "npx"
args = ["@lpextend/mcp-server"]

Gemini CLI

Add to your settings.json:

{
  "mcpServers": {
    "liveperson": {
      "command": "npx",
      "args": ["@lpextend/mcp-server"]
    }
  }
}

VS Code (GitHub Copilot)

Add to .vscode/mcp.json in your project:

{
  "servers": {
    "liveperson": {
      "type": "stdio",
      "command": "npx",
      "args": ["@lpextend/mcp-server"]
    }
  }
}

Cursor / Windsurf

Both support MCP servers through their settings UI. Add a new MCP server with:

• Command: npx
• Args: @lpextend/mcp-server

Credential Loading

The server loads credentials in this priority order:

1. System environment variables (highest priority)
2. Project-level .env file (in the lp-extend-mcp/ directory)
3. User-level ~/.config/lp-mcp/.env (from --setup)

This means you can override the global config per-project by placing a .env file in the project directory.

Optional: Web Crawl (Playwright)

The web_crawl tool extracts text content from any webpage — including JavaScript-rendered SPAs that standard HTTP fetching can't handle. It uses headless Chromium via Playwright to fully render pages before extraction.

This is especially useful for the Website → Knowledge Base workflow: crawl a brand's website, extract content, and feed it into composite_kai_kb to create a fully populated knowledge base.

To enable it, install Playwright after the main install:

npm install playwright
npx playwright install chromium

If Playwright is not installed, the tool is silently omitted — all other tools work normally.

Available Tools

62 tools across 10 modules (+ 1 optional). Each tool uses an action parameter to select operations. See TOOLS.md for the full reference with all actions and parameters.

| Module | Tools | What they do | |--------|-------|--------------| | Account Config | ac_skills, ac_users, ac_lobs, ac_lookup, ac_entry_points, ac_windows, ac_campaigns, ac_engagements | CRUD for all account resources | | Account Config (Extended) | ac_workdays, ac_special_occasions, ac_auto_messages, ac_predefined_content, ac_predefined_categories, ac_audit_trail, ac_agent_status, ac_shift_status, ac_agent_survey | Working hours, holidays, canned responses, audit trail, agent status | | Conversation Builder | cb_bots, cb_deploy, cb_dialogs, cb_interactions, cb_integrations, cb_global_functions, cb_lookup, cb_logs | Bot management, deployment, dialog building | | AI Studio | ai_flows, ai_prompts, ai_conversations, ai_categories, ai_knowledge, ai_simulations, ai_evaluation, ai_routing, ai_lookup | Flow CRUD, testing, evaluation, routing | | Knowledge AI | kai_knowledgebases, kai_articles | KB and article management | | Conversations | conv_manage | Search, transcript, close, close all, transfer, send message | | Auth | auth_tokens, auth_apps | OAuth tokens, app installations | | Composites | composite_campaign_stack, composite_kai_kb, composite_web_crawl_to_kb, composite_wire_cb_bot, composite_deploy_bot, cb_bot_health_check, workflow_create_demo, composite_bot_summary, composite_flow_summary, composite_bot_audit, composite_teardown_demo, composite_cleanup_campaigns, composite_describe_flow, composite_describe_bot, composite_flow_audit, composite_optimise_kb, composite_route_test_suite, composite_backup_account | Multi-step workflows, summaries, audits, optimisation, testing, backup, and teardown | | Demo Wizard | demo_get_creation_plan, demo_list_available_bot_users, demo_get_component_info, demo_list_flow_types | Demo creation planning | | Web Crawl (optional) | web_crawl | Extract content from any webpage including SPAs (requires Playwright) |

Usage

Once connected, just talk to your AI assistant in natural language. The MCP server gives it the tools — you describe what you want done.

Exploring your account

Start by asking questions about what's in your account:

"Show me all my skills"
"List all bots and their running status"
"What knowledge bases do I have?"
"Show me all bot users and which bots they're assigned to"
"What profiles are available in my account?"

Website to Knowledge Base

One of the most powerful workflows — turn any website into a KAI knowledge base ready for bot consumption:

"Review https://brand.com/help and create a knowledge base called 'Brand FAQ'
 with articles covering their main help topics"

The AI will:

1. Fetch and analyse the website pages
2. Extract key topics and content
3. Create a new knowledge base
4. Generate articles with properly formatted titles (< 300 chars), summaries (< 1,000 chars), and details (< 10,000 chars)
5. Batch-create all articles in one go

This is particularly useful for demo prep — quickly spin up a realistic KB from a prospect's website.

Bot management

Understand and manage your Conversation Builder bots:

"Which user account is the 'Sales Bot' using, and what skill is it assigned to?"
"The FAQ Bot is stopped — start it"
"Assign bot user 'bot-agent-02' to the Support Bot"
"Show me all bots in the 'Production' bot group"

Conversation operations

Manage live and historical conversations — especially useful for demo cleanup:

"Close all open conversations on the 'Test' skill"
"Transfer conversation abc-123-def to the 'VIP Support' skill"
"Show me the transcript for conversation abc-123-def"
"Find all open conversations from the last 24 hours"

Account setup and configuration

Set up skills, users, and structure for demos or new deployments:

"Create skills called 'Sales', 'Support', and 'VIP'"
"Show me all users with the 'Bot' type"
"What agent groups exist and who's in them?"
"List all lines of business"

Chaining actions

The real power is in combining multiple steps with a single instruction:

"Set up a demo account: create skills 'Sales', 'Support', and 'VIP',
 then show me available bot users I can assign to each"

"Review https://acme.com/faq, create a KB called 'Acme FAQ',
 then check if the FAQ Bot is running and start it if it's not"

"Find all open conversations on the 'Test' skill, close them,
 then show me the 'Test' skill details to confirm it's clean"

Tips

• Be specific with names — say "the 'Sales Bot'" not just "the bot", especially if you have multiple
• Ask questions first — if you're not sure what's in your account, ask the AI to list things before making changes
• Use it for auditing — "Show me which bots are stopped" or "Which bot users don't have any bots assigned?"
• Destructive actions get confirmed — the AI will confirm before closing conversations, stopping bots, etc.

Authentication

The server authenticates using LP's Application Login API (bot user with API keys). It handles:

• Bearer token acquisition and caching
• Automatic token refresh before expiry
• Retry on 401 (expired token)
• CB/KAI 2-step auth flow (bearer → apiAccessToken + organizationId)

All transparent — you just use the tools.

Development

If you want to work on the MCP server itself (add tools, fix bugs, etc.):

git clone https://github.com/lpdb2024/lp-extend-mcp.git
cd lp-extend-mcp
npm install
copy .env.example .env  # Add your credentials
npm run build

Then add a .mcp.json file in the root of whatever project you open in your AI tool. This tells the tool how to start the server:

{
  "mcpServers": {
    "liveperson": {
      "command": "node",
      "args": ["C:/path/to/lp-extend-mcp/dist/index.js"]
    }
  }
}

Or register it globally (Claude Code) so it's available in every project:

claude mcp add --scope user liveperson -- node C:\path\to\lp-extend-mcp\dist\index.js

You don't run the server yourself — your AI tool spawns it automatically when it needs to call a tool. The config just tells it which command to run.

To test the server manually without an AI tool, use the MCP Inspector:

npm run inspect

License

MIT