crm-onboarding

CRM Onboarding

You are helping a user get started with Open Tooling CRM — an open-source, local-first, headless CRM designed for AI agents. CRM is the first module in the Open Tooling family.

When to Use This Skill

• User says "how do I get started with the CRM" or similar
• User asks Claude to "use Open Tooling CRM" or "start using the CRM"
• User just installed this plugin and wants to try it out
• User asks about setup, configuration, or first steps

Onboarding Flow

Step 1: Check if CRM is already connected

Try calling the search_entities MCP tool with no filters. This is the fastest way to know if the CRM is set up and connected.

• If it works → skip to "CRM is Connected" below.
• If the tool is not available or fails → the CRM needs to be set up. Continue to Step 2.

Step 2: Install the CRM locally

The user needs to install the CRM on their local machine (not in this sandbox/session).

IMPORTANT: Do NOT run setup commands inside Cowork's sandbox or session terminal. Cowork sessions are ephemeral — files installed there disappear when the session ends. The user must run these commands in their own local terminal (e.g., iTerm, Terminal.app, Windows Terminal).

Detect the user's OS and show the appropriate commands.

macOS / Linux — say exactly this:

Open your local terminal (not this chat) and run:

git clone https://github.com/Attri-Inc/open-tooling.git ~/open-tooling
cd ~/open-tooling/crm
npm install && cp .env.example .env
npm run seed

Let me know when that's done and I'll configure the MCP connection for you.

Windows — say exactly this:

Open PowerShell or Windows Terminal (not this chat) and run:

git clone https://github.com/Attri-Inc/open-tooling.git $HOME\open-tooling
cd $HOME\open-tooling\crm
npm install; copy .env.example .env
npm run seed

Let me know when that's done and I'll configure the MCP connection for you.

Do NOT add anything else. Wait for the user to confirm.

Step 3: Configure the MCP connection

Once the user confirms the install is done, say:

I can add the CRM MCP server to your Claude Desktop config so it's available in every Cowork session. Want me to do that?

If the user agrees, update the config file.

CRITICAL: This file is on the user's LOCAL machine, NOT in the Cowork sandbox. The path depends on the OS:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json (typically C:\Users\<username>\AppData\Roaming\Claude\claude_desktop_config.json)

You MUST access this file on the host filesystem — request access to the appropriate folder if prompted. Do NOT look for it inside the sandbox working directory. This is the same file the user would open via Claude Desktop → Settings → Developer → Edit Config.

Steps:

1. Request access to the folder and read the file at the correct local path for the user's OS
2. Parse the existing JSON (it may have preferences and other keys — preserve them all)
3. If there's no mcpServers key, create it. Add the open-tooling-crm entry:

   {
     "open-tooling-crm": {
       "command": "npx",
       "args": ["tsx", "<HOME>/open-tooling/crm/src/mcp.ts"],
       "env": {
         "CRM_DB_PATH": "<HOME>/open-tooling/crm/data/crm.db"
       }
     }
   }
   

   Replace <HOME> with the user's actual home directory path (e.g., /Users/username on macOS, C:/Users/username on Windows). Use forward slashes on all platforms. Determine this from the config file path or the CRM install path.
4. Write the updated JSON back. Do not overwrite other keys — only add/update mcpServers["open-tooling-crm"].
5. Say:

Done! I've added the CRM MCP server to your Claude Desktop config. Now restart Claude Desktop and start a new conversation — I'll have full access to your CRM data.

If the user declines, show the manual config instructions as a fallback.

CRITICAL — you MUST follow these rules:

• Do NOT run /crm-setup or any setup commands inside this session. The sandbox is ephemeral and files will be lost.
• There is NO step about running npm run dev. The MCP server is launched directly by Claude Desktop.
• There is NO step about syncing, updating, or reloading the plugin.
• Do NOT search the MCP registry or suggest third-party CRM connectors (Zoho, HubSpot, Attio, etc.).
• Do NOT add extra steps, numbered lists, explanations of MCP, or architecture overviews.

---

CRM is Connected

Once search_entities works, walk the user through what they can do:

1. Search contacts: Use search_entities with type: "contact"
2. View the deal pipeline: Use search_entities with type: "deal"
3. Explore relationships: Use traverse_graph on a contact to see their company, deals, interactions
4. Read a brief: Use list_briefs then get_brief to see a summary backed by evidence
5. Drill into evidence: From a brief, follow observation IDs to get_observation, then get_artifact for the raw source

If the database was seeded, jump straight in and show the user their data.

Key Concepts (explain if asked)

• Entities: Typed records (contact, company, deal, interaction, task, agent) with JSON properties
• Relationships: Directed edges (EMPLOYED_AT, ASSOCIATED_WITH, etc.)
• Memory Layer: Artifacts → Observations → Briefs → Conflicts (evidence chain)
• Event Ledger: Append-only audit trail for every mutation
• Progressive Retrieval: Start with briefs, drill to observations, then artifacts

Key Principles

• Always use the MCP tools (not REST API) when working through Claude
• The memory layer is the differentiator — every claim has receipts
• Progressive retrieval keeps context lean while maintaining full traceability