AI assistants (MCP)

SignSealer is a remote MCP server at https://api.signsealer.com/mcp. Connect it to Claude, ChatGPT, Codex, Claude Code, Cursor or VS Code, and ask: "send the rental agreement to Ada Lovelace", "who still hasn't signed tomorrow's waivers?", "remind them".

The Model Context Protocol is how an assistant is given tools. Ours are the API: each tool makes the same /v1 call a key would, with the same scopes, limits and answers, so an assistant can do exactly what the person who connected it allowed and nothing the API would refuse.

Connect it

Every client below signs in the same way: it opens SignSealer's sign-in, you choose the business, see what it is asking for, and press Connect. There is no key to copy. You need the permission to manage API keys in that business, which owners and admins have (Teams, roles and sign-in). You can disconnect it at any time under Integrations → Connected apps, and it stops that minute.

Claude (claude.ai and Claude Desktop)

Settings → Connectors → Add custom connector. Name it SignSealer, give the URL https://api.signsealer.com/mcp, and press Connect.

ChatGPT

Settings → Apps & Connectors. With developer mode on, create a connector with the URL https://api.signsealer.com/mcp and OAuth as its authentication.

Claude Code

claude mcp add --transport http signsealer https://api.signsealer.com/mcp

Then /mcp in a session to sign in.

Codex

In ~/.codex/config.toml:

[mcp_servers.signsealer]
url = "https://api.signsealer.com/mcp"

codex mcp login signsealer

Cursor

In ~/.cursor/mcp.json for every project, or .cursor/mcp.json in one:

{
  "mcpServers": {
    "signsealer": { "url": "https://api.signsealer.com/mcp" }
  }
}

Cursor signs in when it first connects. To add it in one step instead, open this link in a browser on the computer Cursor is on:

cursor://anysphere.cursor-deeplink/mcp/install?name=signsealer&config=eyJ1cmwiOiJodHRwczovL2FwaS5zaWduc2VhbGVyLmNvbS9tY3AifQ==

VS Code

In .vscode/mcp.json in a workspace, or in your own mcp.json (the command MCP: Open User Configuration):

{
  "servers": {
    "signsealer": { "type": "http", "url": "https://api.signsealer.com/mcp" }
  }
}

VS Code shows Start above the server, and signs in when it starts.

Anything else

A client that speaks MCP over HTTP finds the sign-in by itself from the URL:

{
  "mcpServers": {
    "signsealer": { "url": "https://api.signsealer.com/mcp" }
  }
}

One that cannot sign in can send an API key instead, from Integrations → API keys with signing:read and signing:write, as a header:

{
  "mcpServers": {
    "signsealer": {
      "url": "https://api.signsealer.com/mcp",
      "headers": { "Authorization": "Bearer ${env:SIGNSEALER_KEY}" }
    }
  }
}

Keep the key in the environment, not in a file you commit. A key that holds only signing:read gives an assistant that can look but not send.

What it can do

The four that change something are marked as such to the client, which asks you before it runs one. The server also tells the assistant to confirm the template, the values and every recipient with you before it sends anything, and never to send to an address you did not give it.

Sending takes an idempotency_key the assistant makes up for that one send (reminding and voiding take one too, if given). An assistant whose connection drops calls again with the same key and is answered with what the first call did, so a retry never sends a document twice. Reading a document gives its status and its signers; the text, the signers' answers and the event trail, with its network addresses and devices, come only when asked for.

What you are asked to allow

An assistant asks for signing:read, signing:send, signing:remind, signing:links, signing:void. Reading documents, templates and who has signed is given when you connect it. Each of the four that reaches people or stops a link is a box on SignSealer's sign-in page, and none is ticked until you tick it:

• Send documents for signature (signing:send)
• Remind a signer who has not signed (signing:remind)
• Make a fresh signing link (signing:links)
• Void a document that is not finished (signing:void)

An assistant that only needs to tell you who has signed gets no box ticked. One asked to send a document is refused if Send documents for signature was left clear, and says so: nothing is sent. To change what it may do, connect it again and tick the boxes you want; what you tick replaces what it had, at once, even for the sign-in it already holds.

It is never given more than you hold yourself, and never the scopes that manage templates, webhooks, workflows or keys. An API key with signing:write holds all four.

The sign-in page says Not checked by SignSealer for an app that introduced itself, which every assistant does, and names where your browser goes back to afterwards: the assistant's own site, or an app on this computer for Codex and Claude Code. Connect it only when you started the connection yourself, just now.

For client builders

Building on the API directly instead? The quickstart and the Node client are the way in, and llms-full.txt is every page of these docs as one text file, for a coding assistant to read.