The Lidia Labs MCP: connect your whole business to Claude Code, Cursor and your favorite AI agent

Imagine opening Claude Code and typing: “how many appointments do I have tomorrow?”, “reply on WhatsApp to Mariana that Thursday works”, “tweak Lidia's system prompt to push this week's 2-for-1”. No dashboard, no window switching, no copy-paste. Starting today that is exactly what the Lidia Labs MCP enables: your entire business — contacts, calendar, notes, the WhatsApp inbox, the agent's config and even your team — available as native tools inside the AI agent you already use every day.

In this guide we cover what MCP is, everything our server exposes, how to create your API key with granular permissions (resource by resource, read or write), and how to install it step by step in Claude Code, Cursor, Windsurf, Antigravity, GitHub Copilot, Kiro, Codex and Qoder.

What MCP is and why it matters

The Model Context Protocol (MCP) is an open standard created by Anthropic in late 2024 and since adopted by practically the whole industry: OpenAI, Google, Microsoft, Cursor and dozens of other tools. The idea is simple and powerful: instead of every app inventing its own integration with every AI model, MCP defines a common language — a kind of “USB-C port” for artificial intelligence. An MCP server publishes tools; any compatible client discovers and uses them.

For a business this is a game changer. “Integrating AI with my system” used to mean weeks of custom development. With MCP it means pasting a URL and a key. The agent sees your tools, understands what each one does and combines them on its own: it can look up a contact, review their appointment history and book a new one in a single conversation.

Everything the Lidia Labs MCP exposes

Our server lives at https://lidialabs.com/api/mcp and uses the protocol's official HTTP transport, authenticated with your API key. It doesn't stop at the CRM: it covers almost everything you'd do in the dashboard, organized into ten resources.

• Contacts — search by name, phone or email; create, edit and delete, with per-vertical custom fields.
• Calendar — list appointments by date range, status or contact; book, reschedule and cancel.
• Notes — free-form notes attached to a contact or an appointment.
• Agent (system prompt) — read and update Lidia's instructions, her model and whether she's active.
• Agent images — add to the photo catalog Lidia sends (by file or by URL) and delete them.
• WhatsApp inbox — list conversations and messages, reply on WhatsApp, and close or reopen chats.
• WhatsApp connection — check the number's status and disconnect it.
• Members — list the team, invite, change roles and permissions, and remove members.
• Organization — read and edit org and business settings (name, timezone, locale, currency).
• API keys — list, create and revoke your own keys from the API.

Each operation is a clearly named tool (get_business_info, list_contacts, send_whatsapp_message, update_agent_config, invite_member…) that the agent discovers and combines on its own. And everything is isolated per business: the API key belongs to one specific business in your organization and only sees that business's data. If you run several locations or brands, each one gets its own keys.

Granular permissions: you decide what each key can touch

This is the key difference. An API key isn't all-or-nothing. Each of the ten resources has two independent switches — read and write — so you shape exactly the access you need.

• A reporting dashboard: read-only on calendar and contacts, nothing else. Even if the agent wanted to, it can't change a single record.
• A WhatsApp assistant: read and write on the inbox, but no access to members or the organization.
• An automation that tunes the campaign: write on the agent's system prompt and its images, with no CRM access.
• Team provisioning: write on members to onboard people, and nothing more.

Tools the key isn't allowed to use don't even show up in the agent's list. And to keep it simple, there are two one-click presets: «Read only» (read on everything) and «Full access» (everything).

Create your API key in two minutes

Keys are managed from the dashboard under Settings → Connections:

• Go to lidialabs.com and open Settings → Connections.
• Click “Create API key” and give it a descriptive name (say, “Fede's Claude Code”).
• Flip the read and/or write switches per resource, or use a preset. It starts with read on the core CRM (contacts, calendar and notes); sensitive resources like members or the inbox stay off until you turn them on.
• Copy the key — it starts with lidia_sk_ and, for security, it is shown only once.

Who can create keys? Only the organization owner, or members you grant the “Manage external connections” permission from the Team tab.

Install it in your AI agent

The pattern is always the same: point your client at https://lidialabs.com/api/mcp and send your key in the Authorization header as a Bearer token. Here is what it looks like in each client (replace YOUR_API_KEY with your key):

• Claude Code — one line in your terminal: claude mcp add --transport http lidia https://lidialabs.com/api/mcp --header "Authorization: Bearer YOUR_API_KEY".
• Cursor — add the server in .cursor/mcp.json with url https://lidialabs.com/api/mcp and the authorization header inside headers.
• Windsurf — in ~/.codeium/windsurf/mcp_config.json declare the server with serverUrl and the same header.
• Antigravity — open Manage MCP Servers, edit mcp_config.json and add the server with serverUrl and headers.
• GitHub Copilot (VS Code) — create .vscode/mcp.json with type "http", the server url and the authorization header.
• Kiro — in .kiro/settings/mcp.json use the mcp-remote bridge: command npx with args [-y, mcp-remote, https://lidialabs.com/api/mcp, --header, Authorization: Bearer YOUR_API_KEY].
• Codex — from the terminal: codex mcp add lidia -- npx -y mcp-remote https://lidialabs.com/api/mcp --header "Authorization: Bearer YOUR_API_KEY".
• Qoder — in the MCP settings add the server with url and headers, just like Cursor.

In every case, restart the client and you'll see Lidia's tools available. Ask the agent to call get_business_info as a smoke test: it should answer with your business name.

So the agent understands everything without you explaining anything, the MCP sends full instructions on connect. And if you want to go further, we publish ready-to-install guides: a Claude Code skill (curl https://lidialabs.com/mcp/claude-code/SKILL.md into ~/.claude/skills/), an AGENTS.md for Codex and Antigravity, and equivalent rules for Cursor, Windsurf, Copilot, Kiro and Qoder. The full walkthrough is at https://lidialabs.com/mcp and the complete agent guide at https://lidialabs.com/mcp/lidia-labs.md.

What you can ask once connected

• Your day at a glance: “what appointments do I have today and which ones are still unconfirmed?”.
• Reply for you: “tell Laura on WhatsApp that Thursday at 5 works, then close the conversation”.
• Capture leads on the fly: paste a list of names and phone numbers and ask it to create them as contacts with their source.
• Tune Lidia: “update the system prompt to mention this week's 2-for-1” or “add this photo of the shop to the catalog”.
• Manage the team: “invite ana@email.com as a member” or “give Pedro the external-connections permission”.
• Adjust the organization: “change the business timezone to America/Monterrey”.
• Analyze: “list this month's cancelled appointments and tell me which contacts we should win back”.

The magic is in the combination: the agent chains tools on its own. A “prep my Monday” can turn into listing appointments, reviewing each contact's notes, and leaving confirmation messages ready in the inbox — all in one pass.

There's a REST API too

The MCP shares credentials and logic with our public REST API at https://lidialabs.com/api/v1. If classic integrations are your thing — a homemade Zapier, a reporting script, your own system — you get the same ten resources over HTTP: /contacts, /appointments, /notes, /agent, /agent/images, /inbox, /whatsapp, /members, /organization and /keys, plus /me to verify your key. You authenticate the same way, with Authorization: Bearer and your lidia_sk_. Errors return consistent JSON with a code and a message, and pagination uses limit and offset.

Security: how we protect your data

• Your key is never stored in plain text: we keep only its SHA-256 hash, like serious API providers do, and the full token is shown only once.
• Truly least-privilege: every resource has separate read and write, and out-of-scope tools don't even appear in the agent's list.
• No privilege escalation: a key that can create other keys cannot grant permissions it doesn't hold itself, so a leaked key can't widen its own reach.
• Per-business isolation: each key sees only the business it belongs to, never other organizations or locations.
• Safe team management: managing members via the API can't leave the organization without owners.
• Guarded image uploads: when fetching an image by URL we block internal addresses so nobody can point the API at our own network.
• Instant revocation: revoke any key from the dashboard and the integrations using it stop working right away.
• Team control: only the owner — or members with the explicit external-connections permission — can create or revoke keys.

The most useful AI isn't the one that knows the most — it's the one that can act on your real data, with permissions you control.

The Lidia Labs MCP is available on every plan today. Create your first key under Settings → Connections and plug your business into the future — it takes less time than brewing a coffee.

Sources

• Model Context Protocol — https://modelcontextprotocol.io
• Anthropic, Introducing the Model Context Protocol — https://www.anthropic.com/news/model-context-protocol
• Claude Docs, Connect Claude Code to tools via MCP — https://docs.claude.com/en/docs/claude-code/mcp
• Cursor Docs, Model Context Protocol — https://docs.cursor.com/context/model-context-protocol
• Visual Studio Code Docs, MCP servers — https://code.visualstudio.com/docs/copilot/chat/mcp-servers