MCP Server

Lotics speaks the Model Context Protocol, so the AI assistant you already use can read and write your workspace directly. Connect ChatGPT, Claude Desktop, Claude.ai, or Claude Code once, then ask questions in plain language: "which shipments are missing paperwork", "create a quote for this customer", "how many containers cleared customs last month".

The connector acts as you. It reaches exactly the tables and apps your account reaches — no more, and nothing that would not already appear on your own screens.

Endpoint

https://api.lotics.ai/mcp

One endpoint, two ways to authenticate. Which one you use depends on the client.

Connect ChatGPT

1. In ChatGPT, open Settings → Connectors → Add custom connector (available on Plus, Pro, Team, Enterprise and Edu plans — your workspace admin may need to enable it first).
2. Name it Lotics and set the server URL to https://api.lotics.ai/mcp.
3. Choose OAuth as the authentication method and save.
4. ChatGPT opens a Lotics sign-in page. Sign in, then pick the organization and workspace this connector should use, and approve.

Once connected, start a chat and ask ChatGPT to use Lotics. It can list your tables, query records, run your apps' saved queries, trigger the actions those apps expose, and work with the services your admin has opened to it.

Connect Claude Desktop or Claude.ai

1. Open Settings → Connectors → Add custom connector.
2. Set the remote MCP server URL to https://api.lotics.ai/mcp.
3. Claude opens a Lotics sign-in page. Sign in, pick the organization and workspace, and approve.

The same connector works in the Claude desktop app, on claude.ai, and in the Claude mobile apps.

Connect Claude Code

Claude Code uses an API key rather than a browser sign-in. An admin creates one in Lotics under Settings → API keys, then:

claude mcp add --transport http lotics https://api.lotics.ai/mcp \
  --header "Authorization: Bearer ltk_your_key_here"

If your organization has more than one workspace, add a second header naming the one you want:

  --header "X-Workspace-Id: wsp_your_workspace_id"

Connect any other MCP client

Point the client at https://api.lotics.ai/mcp over streamable HTTP and send your API key as Authorization: Bearer ltk_.... Clients that support OAuth 2.1 with dynamic client registration can use the browser sign-in instead — nothing needs to be registered with us in advance.

A machine-readable descriptor of this server — its name, transport, endpoint and headers — is published at https://lotics.ai/.well-known/mcp.json, for a client or registry that discovers servers rather than being handed a URL.

Choosing a workspace

A connector serves one workspace at a time.

• Browser sign-in (ChatGPT, Claude): you pick the organization and workspace on the approval page. To switch, re-connect the connector and pick a different one — your latest choice replaces the previous.
• API key: send an X-Workspace-Id header. With a single-workspace organization you can leave it out.

What the connector can do

Your tables. List tables, read and search records, create, update or delete records, and create, change or delete a table itself — for every table you have been given access to. Tables you have not been shared are not listed and cannot be read, and where a table restricts rows or hides fields from you, the connector sees the same restricted version you do.

Your apps. Ask the assistant for your apps, then for what one of them can do, and it can run that app's saved queries and actions — the same ones its own buttons and screens use. This works even when you have no direct access to the tables behind the app, exactly as it does when you open the app yourself: an app answers through the queries its author wrote, which fix the rows and columns you get back.

Your knowledge. Browse the document list, search it, read a page, and write one — the handbooks, tariffs, and procedures your workspace keeps, limited to the ones shared with you. You can create a document, revise one, label it, and remove it, so a corpus can be built from wherever you already work. Retrieval matches text literally, so a document answers best when it contains the words people actually ask with.

Documents. Generate a filled Excel, Word, or PDF from one of your templates — the same templates your workflows use — and get back a file. Ask what templates exist and what fields one expects.

Connected services. Where a workspace admin has allowed it, the assistant can work with a service connected to the workspace — stock and sales from KiotViet, master data from MISA, a Lark Base, a FedEx rate. It asks which services are open to it, then reads, and where the admin chose read and write it can also create, change or cancel things there, such as a KiotViet order. Admins set this per connection: the connect form asks whether assistants may use it, with reading switched on where the service has anything to read and writing never on until chosen, and Settings → Connections changes it later.

Your files. Ask what a document says and the assistant reads it back to you — Word, Excel, CSV and text arrive as content it can work with. Ask for the file itself and you get a link that opens for an hour, no login needed.

Files go the other way too. Give the assistant a public web address and the file lands in your workspace, ready to attach to a record. For a file the assistant is holding rather than linking to — one it just produced, or one you attached to the conversation — it asks Lotics for an upload address and sends the file there directly, up to 20 MB. The same file sent twice is stored once.

In ChatGPT, a file you attach to the chat can be saved into your workspace without any of this.

If your assistant runs in a sandbox, allow it to reach api.lotics.ai. Assistants that execute code do so inside a sandbox whose outbound network is restricted to an approved list of addresses. Uploading a file the assistant holds is an ordinary web request to Lotics, so it needs api.lotics.ai on that list — otherwise the assistant reports that the connection was refused. In Claude, an owner of a Team or Enterprise account sets this under Admin settings → Capabilities. Reading files, and everything else the connector does, is unaffected.

Anything the assistant writes is recorded as your change, appears in the record's history, and fires the same automations a change made in the Lotics interface would.

What it cannot do

• Reach a table, an app, a document, or a workspace you were not given access to. Losing access takes effect on the next request, not at the next sign-in.
• Turn app access into table access. Being able to use an app never lets the connector read that app's tables directly.
• Change permissions, manage members, create API keys, or touch billing. A browser-signed-in connector is confined to this endpoint and cannot reach the rest of the API at all.
• Touch a connected service an admin has not opened to it. Reading needs at least read only on that connection, and a booking, an accounting voucher, an e-invoice, a post or a KiotViet order needs read and write. Below that level, those still live in a Lotics workflow, which the assistant can run.
• Build things. Editing workflows, automations, apps, or document templates stays in Lotics, where the guidance for getting them right lives.

If your assistant reports that something is not permitted, that is a sharing question — ask whoever owns the table or app to share it with you, and the connector picks the change up immediately.

Security notes

• Approve deliberately. Anything you connect can read what you can read. Connect assistants you trust, in workspaces you mean to expose.
• Revoke from the client. Removing the connector in ChatGPT or Claude stops its access. To cut off an API-key client, disable the key under Settings → API keys.
• Confirm writes. Assistants act on their reading of your request. For anything irreversible, ask it to show you what it plans to change first — and consider putting an approval step in the workflow itself, which applies no matter who or what triggers it.

Related

• REST API — the same capabilities as plain HTTP endpoints
• CLI & SDK — for scripting and coding agents working in a terminal
• Permissions — how access to tables and apps is granted
• Apps — what an app's saved queries and actions are