- a remote MCP server at
https://api.sendforsign.com/mcpthat any MCP client (Claude, ChatGPT, Cursor, the OpenAI Agents SDK, your own agent) can connect to, and - the REST API at
https://api.sendforsign.com/apithat the MCP server itself calls, for when you want deterministic code around the agent.
MCP quickstart
Connect Claude Code, Claude Desktop, Cursor or an OpenAI agent in two minutes.
Tools reference
Every MCP tool, what it writes, and which one emails people.
Agent modes and accountability
Read-only, draft and send modes, confirmations, domain allowlists and the audit log.
Recipes
Template → fill → send, embedded signing without email, multi-tenant, webhooks.
Why Sendforsign, if other e-signature tools also ship MCP
An MCP server is table stakes; most e-signature vendors have one. What matters for a product that builds agents is what happens around the signature:1. Templates with field roles
Each field (placeholder) in a template has a filled by role: the owner (your agent), any recipient, or one specific recipient. The agent reads the template, fills only its own fields and leaves the counterparty’s fields to the counterparty. It does not have to guess, and it cannot overwrite what a signer is supposed to enter. See placeholderfillingType.
2. Embedded, white-label, multi-tenant
- Every customer of your product gets its own space (a
clientKey, and/or auserKeyper end user) with its own templates and contracts. The agent works in the right tenant because you set it in a header, not the model. - Recipients get a signing link (
https://app.sendforsign.com/sharing/{recipientKey}) that you can show in your own UI, chat or app, with no Sendforsign email at all. - Emails, when you do send them, carry your branding, and webhooks tell your backend when a contract is fully signed.
- The React components let a human review and edit what the agent drafted, inside your product.
3. Agent accountability
The integrator, not the agent, decides what an agent may do:- Modes:
read_only,draft(prepare everything, email nobody) orsend. - Explicit confirmation: tools that email people or archive contracts refuse to run without
confirm: true, which the agent is told to set only after a human approves. - Recipient domain allowlist: an agent cannot send a contract outside the domains you list.
- Audit log: every write by an agent is recorded with the agent’s name, the person or company it acted for, the mode, the arguments and the outcome, in a hash-chained log.
Who this is for
- Vertical SaaS with an agent (creative and event businesses, staffing, real estate, logistics, agencies, HR, insurance): your agent drafts your customer’s contract from your customer’s template, sends it or hands the link to your UI, and your backend reacts to the signature.
- Legal-AI startups that draft or negotiate but don’t want to build execution: upload the final text, add signers, send, get the signed PDF.
- People using Claude, ChatGPT or Cursor directly: “make an NDA from our template and send it to Anna”.
For LLMs and coding agents
/llms.txt— an index of these docs for language models./llms-full.txt— the API and MCP reference in one file.- Every page is also available as Markdown: add
.mdto its URL. - Agent skill — a
SKILL.mdthat teaches Claude Code, Codex and other agents the recipes and the pitfalls.