---
title: "Tamloot — MCP Server for Developers"
description: "Connect Claude, ChatGPT, Cursor, or any MCP client to your Tamloot data: look things up and make small edits, secured with OAuth and scoped to your account."
canonical: "https://tamloot.cc/en/docs/developers/mcp-server"
last-updated: "2026-10-20"
---

# MCP Server for Developers

Tamloot runs an OAuth-secured [MCP (Model Context Protocol)](https://modelcontextprotocol.io) server. Connect Claude, ChatGPT, Cursor, or any other MCP client, and it can work with your Tamloot data: look up the people you work with, your sessions, notes, and transcripts, and add or update contacts and notes.

Your AI tool sees what your account sees, and nothing more. There is no API key to manage.

## Connect

The endpoint is `https://mcp.tamloot.cc/mcp` (Streamable HTTP, OAuth 2.1).

**Claude Code**

```bash
claude mcp add --transport http tamloot https://mcp.tamloot.cc/mcp
```

**Claude on the web and desktop, ChatGPT:** add a custom connector and paste the endpoint above.

**Any client that uses a JSON config file**

```json
{
  "mcpServers": {
    "tamloot": {
      "type": "http",
      "url": "https://mcp.tamloot.cc/mcp"
    }
  }
}
```

The first time you connect, your client opens a browser window and asks you to log in with your Tamloot account. Your client handles the rest.

---

## What your AI tool can do

Five tools read. Three make small changes. A connected AI tool can create a contact, update a contact's details, and add a note to a contact. It can't delete anything, and it can't change anything else in your Tamloot account.

| Tool | What it does |
|------|--------------|
| `get_contact` | Find a person by ID, or by name, email, or phone. Returns their notes, recent sessions, and upcoming sessions. Archived contacts are included and marked. |
| `search_sessions` | Search your sessions by keyword, contact, or date range. With no keyword, it lists your most recent sessions. |
| `get_session` | Get one session with its details and notes. |
| `get_transcript` | Read the transcript of a session, in parts. |
| `search_notes` | Search the notes on your sessions and contacts by keyword. |
| `create_contact` | Create a contact with a full name and, optionally, an email and phone number. Refused if a contact with that email already exists. |
| `update_contact` | Change a contact's name, email, or phone. Fields you leave out stay as they are. Pass `null` to clear an email or phone. |
| `add_contact_note` | Add a note (up to 10,000 characters) to a contact. It appears as your own note. |

The five reading tools tell the client that they only read, so clients that support it can let you approve them once instead of on every call. `update_contact` tells the client that it overwrites existing details, so clients such as Claude ask you before running it.

### How changes work

- Changes apply to your own contacts, and to organization contacts that you can see, the same as in the Tamloot app.
- Archived contacts can be read but not changed.
- `create_contact` and `add_contact_note` accept an optional `idempotency_key`. If a call is retried with the same key within 24 hours, you get the original result instead of a duplicate. Using the same key for a different request is refused.
- Each change is recorded in your account's audit log.

### Long results come in parts

Searches and transcripts can be long. The server sends them in pages, and your AI tool asks for the next one on its own.

- `search_sessions` and `search_notes` return `has_more` and `next_offset`. Pass `next_offset` back as `offset` for the next page.
- `get_transcript` pages by character. Each part reports `total_chars`, `has_more`, and `next_offset`, and the recording it came from. Parts are 40,000 characters by default, up to 60,000 with `max_chars`, and end on a line break when they can.
- To read the next transcript part, pass back `offset`, `recording_id`, and `total_chars` from the previous part. If the transcript was reprocessed in the meantime, the call is refused instead of returning mixed text, and the client starts again from the beginning.

### Typed results

Every tool returns its result as JSON text and as structured content, and publishes an output schema. Clients that validate output can rely on the shape, and clients that don't still get readable text.

### Built-in guidance

When a client connects, the server sends a short description of how the tools fit together: where IDs come from, and how paging works. Your AI tool uses the tools correctly without extra prompting from you.

## Tips

- **Names in another alphabet.** Search matches names as they are stored. A name saved in Latin letters won't match its Hebrew, Russian, or Arabic spelling, or the reverse. If a search finds nothing, try the other spelling. Your AI tool usually does this by itself.
- **Typos are fine.** Searches tolerate small spelling mistakes.
- **Archived contacts.** `get_contact` still finds them and shows when they were archived.

## Troubleshooting

| You see | What it means |
|---------|---------------|
| "Tamloot couldn't confirm this account's access to AI tool connections" | Tamloot couldn't check your access just then. Try again. If it keeps happening, email [support@tamloot.cc](mailto:support@tamloot.cc). |
| "A contact with this email already exists" | Use `get_contact` to find the person, then update them or add a note instead. |
| "This contact is archived" | Archived contacts can't be changed from an AI tool. Restore the contact in Tamloot first. |
| "That idempotency_key was already used for a different request" | Use a new key, or repeat the original request. |
| A login window opens again | Your client's session expired. Log in again. |
| "This transcript changed since the previous part" | The transcript was reprocessed while it was being read. Ask for it again from the start. |

---

## Data access & privacy

Every request runs under your own account's permissions. A connected AI tool only sees what your account can see in Tamloot, and the database enforces that, not only the server. Direct writes to your data are blocked at the database. The only changes a connected tool can make are the three above, through one audited function. Tamloot keeps an audit record of each tool call. See our [Privacy Policy](/privacy) for how we handle data more broadly.

---

## Which clients work with this

Any MCP client that supports Streamable HTTP transport and OAuth 2.1 works, including:

- Claude on the web and desktop, and Claude Code
- Cursor
- ChatGPT, via its MCP connector support

If your client doesn't support OAuth-secured remote servers yet, check its own documentation for how to add a remote MCP server. The endpoint above and the login flow stay the same.

---

## Discovery

Tamloot's server is listed in the official [MCP Registry](https://registry.modelcontextprotocol.io/v0.1/servers?search=cc.tamloot/mcp) as `cc.tamloot/mcp`, verified against the tamloot.cc domain. Its server card, with the tool list, is at `https://mcp.tamloot.cc/.well-known/mcp/server-card.json`.

---

## Public documentation MCP

To let an AI tool search and read these guides without signing in, add this separate server:

```json
{
  "mcpServers": {
    "tamloot-docs": { "url": "https://tamloot.cc/api/docs/mcp" }
  }
}
```

It provides `list_pages`, `search` (with `locale: "en"` or `"he"`), and `get_page`. It only reads public documentation. Use the account connector above to access your own Tamloot data.

## Where to look next

- [Tamloot](https://tamloot.cc/en)
- [All guides](https://tamloot.cc/en/docs)
- [llms.txt](https://tamloot.cc/docs/llms.txt)
- [Sitemap](https://tamloot.cc/sitemaps/sitemap.xml)
