developer.law for Agents

The four things your agent can do

• ✓Read your organization's state. Company context, matter status, which agreements exist, what is outstanding. Your agent stops asking you questions you already answered.
• ✓Draft from approved templates. Not an ad hoc prompt against a blank page — a draft built from the same lawyer-maintained process a Story attorney would run, populated with your actual deal terms.
• ✓Route counterparties. Send the right document to the right person with role-aware visibility controls, so the client sees what the client should see and your subcontractor sees what your subcontractor should see.
• ✓Escalate to a real attorney. When the client's redline does something your agent should not be deciding about, it hands off to Story counsel instead of guessing confidently.

Mapped to the week you actually have

The four things your agent cannot do

This list is the reason to trust the rest of it. These are hard limits enforced at the platform, not guidelines an agent is asked to follow:

1. 1)Execute a signature. E-signature requires human identity verification. Your agent can get a document all the way to signature-ready. A person signs it.
2. 2)Escalate a draft's visibility. Moving a document from internal to client-visible is a human decision, because it is often a privilege decision.
3. 3)Approve what a counterparty sees or discloses. Counterparty comments and disclosures require human approval.
4. 4)Reach outside your organization. Access is scoped to your org at the middleware layer, so an agent cannot read another organization's data regardless of how it is prompted. Bulk data room export is also off the table.

How it actually connects

Four steps, and if you have configured an MCP server before, you already know three of them:

1. 1)Authenticate — with a Story-issued API key, scoped to your organization.
2. 2)Connect — point your agent at the Story MCP endpoint.
3. 3)Orchestrate — run legal workflows with policy-aware routing.
4. 4)Escalate — bring in Story counsel when a call needs human judgment.

{
  "mcpServers": {
    "story": {
      "url": "https://backend.story.law/mcp/story-agent",
      "type": "http"
    }
  }
}

Prefer plain text? This page is also available as markdown — built for feeding directly to your own agent.

Every action an agent takes is logged in an audit trail that distinguishes programmatic actions from human decisions — which matters later, when someone asks who approved what.