@launchpath-ai/mcp-server 1.0.0

package/skills/deploy-agent/SKILL.md

@@ -0,0 +1,74 @@
+---
+name: deploy-agent
+description: Create, train, and deploy an AI agent for a client's business. Scrapes their website, builds a knowledge base, tests it, and deploys to widget, WhatsApp, or API.
+argument-hint: [business description or website URL]
+allowed-tools: mcp__launchpath__*
+context: fork
+---
+
+You are deploying an AI agent using LaunchPath. Follow this workflow precisely:
+
+## Step 1: Parse the user's input
+
+- If they gave a **website URL**, you'll scrape it for knowledge
+- If they **described a business**, use that as context for agent generation
+- If both, use both
+
+## Step 2: Create the agent
+
+Use `create_agent_from_prompt` with a detailed prompt that includes:
+- The business type and industry
+- What the agent should help with (support, sales, scheduling, etc.)
+- Tone and personality guidelines
+- Any specific instructions from the user
+
+If the user provided a website URL, pass it as `website_url` so the AI can use it for context.
+
+## Step 3: Add knowledge
+
+- If a website URL was provided, call `discover_pages` to find all pages on the domain
+- Then call `scrape_website` for each relevant page (prioritize: about, services, pricing, FAQ, contact)
+- Generate relevant FAQs based on the scraped content and call `add_faq` for each one
+
+## Step 4: Test
+
+- Call `chat_with_agent` with 2-3 test questions relevant to the business
+- Verify the agent responds accurately using the scraped knowledge
+- If responses are poor, add more knowledge or FAQs and re-test
+
+## Step 5: Ask about deployment
+
+Ask the user which channel they want to deploy to:
+
+### Option A: Website Widget (default)
+1. Call `deploy_channel` with `channel_type: "widget"`
+2. Call `set_widget_appearance` to customize colors, theme, avatar, and position
+3. Call `set_widget_content` to set welcome message, conversation starters, and pre-chat form
+4. Call `get_embed_code` to get the ready-to-paste HTML snippet
+
+### Option B: WhatsApp
+Tell the user: "For WhatsApp, we need to set up a campaign. Let me walk you through it."
+Then follow the WhatsApp campaign flow:
+1. Create a client if one doesn't exist: `create_client`
+2. `enable_toolset("campaigns")` then `create_campaign` with `channel_type: "whatsapp"`
+3. Tell the user they must enter Meta credentials in the LaunchPath dashboard before proceeding
+4. Once credentials are saved, use `enable_toolset("whatsapp")` and guide them through templates, contacts, and broadcasts
+
+### Option C: API Channel
+1. Call `deploy_channel` with `channel_type: "api"`
+2. The response includes the public chat endpoint URL and bearer token
+3. Explain how to use the API endpoint to build custom integrations
+
+### Option D: Skip deployment
+If the user doesn't want to deploy yet, skip this step. Remind them they can deploy later.
+
+## Step 6: Present results
+
+Summarize what was created:
+- Agent name and what it does
+- Knowledge base contents (pages scraped, FAQs added)
+- Test conversation results
+- Deployment details (embed code, WhatsApp status, or API endpoint)
+- Link to LaunchPath dashboard for further configuration
+
+Always confirm the agent name and greeting message with the user before deploying.

package/skills/launchpath-guide/SKILL.md

@@ -0,0 +1,60 @@
+---
+name: launchpath-guide
+description: Internal guide for using LaunchPath MCP tools to manage AI agents, clients, and deployments.
+user-invocable: false
+---
+
+# LaunchPath Platform Guide
+
+LaunchPath is an AI agent deployment platform. You have access to it via MCP tools prefixed with `mcp__launchpath__`.
+
+## What You Can Do
+
+- **Create AI agents** from natural language descriptions or blank templates
+- **Train agents** with website content (auto-scraped) and FAQ pairs
+- **Deploy agents** to website widgets, WhatsApp, or API endpoints
+- **Manage clients** — create client accounts and link agents as campaigns
+- **Test agents** via chat to verify quality before going live
+- **Monitor** conversations and analytics
+
+## Common Workflows
+
+### Build an agent for a new client
+1. `create_agent_from_prompt` — describe what the agent should do
+2. `discover_pages` → `scrape_website` — add the client's website content
+3. `add_faq` — add common questions and answers
+4. `create_client` — create the client account
+5. `create_campaign` — link the agent to the client
+6. `deploy_channel` — create a website widget
+7. `get_embed_code` — get the HTML to paste on their site
+8. `chat_with_agent` — test it works correctly
+
+### Update an existing agent
+1. `list_agents` — find the agent
+2. `get_agent` — read current config
+3. `update_agent` — change prompt, greeting, model, etc.
+4. `chat_with_agent` — test the changes
+
+### Check on your business
+1. `list_agents` — see all agents and their status
+2. `list_clients` — see all clients
+3. Enable analytics toolset, then `list_conversations` — check recent activity
+
+## Tips
+
+- Always scrape the client's website before going live — it dramatically improves agent quality
+- Use `chat_with_agent` to test after every change
+- Deploy as widget first — it's the fastest path to a working agent
+- Create the client and campaign before deploying so conversations are properly tracked
+- Use `discover_pages` before `scrape_website` for comprehensive coverage
+- The default model (openai/gpt-4o-mini) is the most cost-effective. Upgrade to claude-sonnet for higher quality.
+
+## Available Toolsets
+
+Default (always loaded): agent management, knowledge, deployment, clients
+On-demand (use `enable_toolset`):
+- **campaigns** — link agents to clients
+- **whatsapp** — templates, broadcasts, contacts, sequences
+- **analytics** — metrics, conversation history
+- **tools-config** — webhooks, APIs, Composio integrations
+- **portal** — client portal invites and branding

package/skills/setup-client/SKILL.md

@@ -0,0 +1,44 @@
+---
+name: setup-client
+description: Create a client account on LaunchPath, link an existing agent as a campaign, and generate deployment details.
+argument-hint: [client name]
+allowed-tools: mcp__launchpath__*
+context: fork
+---
+
+You are setting up a new client on LaunchPath. Follow this workflow:
+
+## Step 1: Gather information
+
+Ask the user for:
+- **Client name** (required)
+- **Client email** (optional)
+- **Client website** (optional)
+- **Which agent to assign** — show the list from `list_agents` and let them pick
+
+## Step 2: Create the client
+
+Call `create_client` with the provided details.
+
+## Step 3: Create the campaign
+
+Call `create_campaign` linking the chosen agent to the new client. Use a descriptive name like "[Client Name] - [Agent Name]".
+
+## Step 4: Deploy
+
+Call `deploy_channel` on the agent with `channel_type: "widget"` to create a website widget.
+Call `get_embed_code` to get the HTML snippet.
+
+## Step 5: Optional portal access
+
+Ask if they want to invite the client to the portal:
+- If yes, call `enable_toolset("portal")` first, then `invite_client_member` with the client's email
+
+## Step 6: Present results
+
+Summarize:
+- Client created with ID
+- Campaign linked (agent → client)
+- Embed code ready to paste
+- Portal invite status
+- Next steps (customize widget, add more knowledge, etc.)

package/skills/whatsapp-campaign/SKILL.md

@@ -0,0 +1,72 @@
+---
+name: whatsapp-campaign
+description: Set up a complete WhatsApp campaign — create templates, import contacts, and launch broadcasts or drip sequences.
+argument-hint: [campaign name or description]
+allowed-tools: mcp__launchpath__*
+context: fork
+---
+
+You are setting up a WhatsApp campaign using LaunchPath. Follow this workflow:
+
+## Step 1: Load required toolsets
+
+Call `enable_toolset("campaigns")` and `enable_toolset("whatsapp")`.
+
+## Step 2: Find or create the campaign
+
+- Call `list_campaigns` to check if a WhatsApp campaign already exists
+- If not, ask the user which agent and client to use (call `list_agents` and `list_clients` to help)
+- Call `create_campaign` with `channel_type: "whatsapp"` to create the campaign
+
+## Step 3: Check WhatsApp credentials
+
+The response from create_campaign will indicate if Meta credentials are needed. If so:
+- Tell the user: "You need to enter your Meta Business credentials in the LaunchPath dashboard before we can proceed."
+- Provide the direct link to the campaign settings page
+- Wait for the user to confirm credentials are saved before continuing
+
+## Step 4: Create message templates
+
+- Ask the user what messages they want to send
+- Call `create_wa_template` for each message template
+- Explain that Meta must approve templates (1-24 hours) before they can be sent
+- Use `list_wa_templates` to check approval status
+
+Template tips:
+- Use {{1}}, {{2}} for variables (e.g., "Hi {{1}}, your appointment is on {{2}}")
+- MARKETING category for promotions, UTILITY for transactional, AUTHENTICATION for login codes
+- Template names must be lowercase with underscores only
+
+## Step 5: Import contacts
+
+- Ask the user for their contact list
+- Call `import_wa_contacts` with the contacts (phone number required, name/email/tags optional)
+- Use tags for segmentation (e.g., "vip", "new-customer")
+
+## Step 6: Choose sending strategy
+
+Ask the user: "Would you like to send a one-time broadcast or set up a drip sequence?"
+
+### Option A: One-time broadcast
+1. Verify templates are approved: `list_wa_templates`
+2. Call `send_wa_broadcast` with:
+   - `contact_ids` for specific contacts, OR `audience_filter` to target by tags/status
+   - `variable_mapping` to personalize messages (e.g., `{ "1": "name" }`)
+   - Optionally `scheduled_for` to schedule for a future time
+
+### Option B: Drip sequence
+1. Call `create_wa_sequence` with timed steps:
+   - Each step has a `delay_hours` (0 for immediate, 24 for next day, etc.)
+   - Use `variable_mapping` per step for personalization
+   - Set `stop_on_reply: true` if the sequence should stop when someone replies
+2. Call `activate_wa_sequence` with `status: "active"` to start the sequence
+3. Call `enroll_wa_contacts` with contact IDs or a filter to add contacts
+
+## Step 7: Present results
+
+Summarize what was set up:
+- Campaign name and linked agent
+- Templates created (and their approval status)
+- Number of contacts imported
+- Broadcast sent or sequence activated
+- Link to dashboard for monitoring