Bring your agent

Your agents can drive PromptEden.

Connect Claude Code, Cursor, ChatGPT agents, or your own orchestration over hosted MCP and the v1 API. Reads are scoped. Writes flow through the same approval and policy system as Eden.

Use MCP or API keys today. OAuth + PKCE is available for provisioned connector clients. CLI npm packaging is coming soon.

Connect MCP API quickstart

// Remote MCP config

{
  "mcpServers": {
    "prompteden": {
      "url": "https://app.prompteden.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ${PROMPTEDEN_API_KEY}"
      }
    }
  }
}

Surfaces

Agent-first access, without bypassing policy.

Hosted MCP

Claude Code, Cursor, ChatGPT agents, and other MCP clients can connect to PromptEden at the hosted MCP endpoint.

https://app.prompteden.com/api/mcp

API keys

Server-side agents can use scoped pe_ keys for REST and MCP bearer-token auth.

Authorization: Bearer pe_...

OAuth + PKCE

Connector clients can use OAuth authorization code with S256 PKCE. Clients are provisioned by PromptEden.

/oauth/authorize -> /oauth/token

Coming soon

CLI

The CLI package is coming soon on npm. Use hosted MCP or API keys today.

@prompteden/cli

Approval over MCP

Agents can propose writes. PromptEden decides when they run.

Every write action your agent proposes goes through the same queue Eden uses in the app. Approve it once, reject it, or turn it into a standing permission with a credit ceiling.

Read tools return data directly when the bearer token has the required scope.

Hosted MCP write tools return approval_required with an approval ID, estimate, and browser deep link.

The human approves in the app or grants a standing permission from the same approval system Eden uses.

The agent polls check_approval until the action is executed, rejected, expired, vetoed, or failed.

Agent result

{
  "status": "approval_required",
  "tool": "content_generate_article",
  "approvalId": 42,
  "deepLinkUrl": "https://app.prompteden.com/approvals/42",
  "estimate": { "credits": 40 }
}

Security model

External agents use Eden's guardrails.

Same approval system

MCP writes map into Eden's app tools and approval queue instead of creating a parallel permission path.

Taint defense on tool output

Untrusted content is marked, disclosed, or withheld before a risky write can become an approvable action.

Scoped identity

API keys and OAuth grants resolve to a team context on the server; agent-supplied tenant fields are not trusted.

Getting started

Bring your agent

Let your agents drive the growth engine

Use hosted MCP or scoped API keys today. Writes still pass through PromptEden approvals, standing permissions, veto windows, and receipts.

Start 7-day free trial Read OAuth guide

Full access for 7 days. Card required, cancel anytime.