Access tokens

An access token connects an AI assistant to Moliri, such as Claude Code, Claude Desktop, Cursor or VS Code. The connection runs over the Model Context Protocol (MCP). The assistant then reads in Moliri with your rights and can suggest field values. It never writes into a field itself: a suggestion is only applied once someone accepts it under Check › AI suggestions.

The page is in the profile under Access tokens. Each person creates their own tokens and sees only their own.

The page

Area	Content
MCP address of this instance	The address the assistant needs, with Copy
List of tokens	For each token the name, the permission, Created …, last used … or never used and, if set, expires …. Expired and revoked tokens are marked as such
Revoke	Blocks the token at once. An assistant that uses it loses access
New token	Top right, opens the dialog for creating one

Creating a token

Setting	Choice
Name	What you will recognise the token by, such as Claude on my laptop
Permission	Read: read the field catalogue, search and read objects. Read and suggest: in addition, suggest field values that you or your team review
Valid for	30 days, 90 days, 1 year or No expiry

New token dialog with the name Claude on my laptop, the permission Read and suggest and a validity of 90 days, at the bottom Cancel and Create token

Accounts that may never change data only get Read. The dialog then says: Your account does not change data, so no suggestions.

Create token shows the token once and only once, together with the ready-made command for Claude Code and the JSON block for other programs, both with the token already filled in. Moliri only stores a check value, not the token itself. If you lose it, revoke it and create a new one.

Setting up in Claude Code

Copy the command under Claude Code and run it in the terminal. It has this form:

SH
claude mcp add --transport http moliri https://…/api/mcp.php --header "Authorization: Bearer mol_…"

After that the server moliri is available in Claude Code. claude mcp list shows whether the connection works, and /mcp does the same in a running session. Without further options the entry applies to the directory in which you ran the command; with --scope user after add it applies everywhere.

Setting up in Cursor

Copy the block under Cursor, VS Code and others (JSON). It has this form:

JSON
{
  "mcpServers": {
    "moliri": {
      "type": "http",
      "url": "https://…/api/mcp.php",
      "headers": { "Authorization": "Bearer mol_…" }
    }
  }
}

In Cursor, add a new server in the settings under MCP. Cursor opens the file mcp.json for this (~/.cursor/mcp.json for all projects, .cursor/mcp.json in the project folder for a single one). Paste the block or, if servers are already listed there, add only the entry moliri under mcpServers, and save. moliri then appears with its tools in the MCP settings.

Other programs that read mcpServers take the same block. Claude Desktop does not support a token in the request header and connects through the helper program mcp-remote. Moliri does not offer sign-in through OAuth, as the connectors in claude.ai or ChatGPT require.

What the assistant may do

Can	Cannot
See the projects you have access to	See more than you see yourself
Read a project’s field catalogue, with the same description as Copy field catalog for AI	Change fields, layouts or settings
Search objects by fields and full text and read individual objects	Write a field value
With Read and suggest: suggest a value with reasoning and evidence, and check what became of its suggestions	Apply a suggestion itself

A suggestion ends up under Check › AI suggestions and as a note at the field. Its origin shows the model the assistant names, otherwise mcp. Only when a person applies it does the value appear in the field and in the history , marked as AI suggestion. A rejected suggestion does not come back, even if the assistant submits the same value again.

The page

Creating a token

Setting up in Claude Code

Setting up in Cursor

What the assistant may do

Further reading