> ## Documentation Index
> Fetch the complete documentation index at: https://docs.earthcare.network/llms.txt
> Use this file to discover all available pages before exploring further.

# Security best practices

> How ECN MCP keeps your workspace safe — org scoping, key handling, and approval gates on anything that sends.

## Authentication & scoping

* **Org-scoped keys.** Every `ecn_mcp_…` key is tied to exactly one organization.
  A key for org A can never read or write org B's data — cross-tenant access is
  rejected.
* **Hashed at rest.** Keys are stored as SHA-256 hashes; the plaintext secret is
  shown only once at creation. If you lose it, create a new key.
* **Revocable.** Revoke any key instantly in **Settings → Connect external AI tools**
  or via `DELETE /api/v1/mcp/keys/<id>/`. Each key tracks `last_used_at`.

## Reads are safe; sends are gated

ECN follows a read/write safety model:

| Action class                                                | Behavior                                                                                                                |
| ----------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------- |
| **Reads** (search, get context, list contacts, balances)    | Run directly and freely.                                                                                                |
| **Workspace writes** (create contact/task, update context)  | Apply to your org's data.                                                                                               |
| **Outreach drafts** (GTM emails, social/outreach campaigns) | Created as approval-gated drafts (`OutreachDraft`, `GtmEmailDraft`) — not sent until a human approves.                  |
| **Direct sends** (`compose_email`)                          | **Send immediately** when the tool is called, via your connected Gmail (Mailgun fallback). Treat these as live actions. |

Most AI-authored content — Context Pack edits and the outreach/GTM pipelines — enters
as `seeded`/`draft` and is **never auto-marked verified**, so a person stays in the
loop before it's published. The exception is direct-send tools like `compose_email`,
which send the moment they're called — only expose them to trusted callers.

<Warning>
  Treat an `ecn_mcp_…` key like a password. Don't commit it to shared repos or
  paste it into project config that syncs publicly. Prefer environment variables.
</Warning>

## What an MCP client can and cannot do

* ✅ Search and read directory + context, manage CRM records, run prospecting,
  draft outreach, check spend.
* ⛔ Bypass credit budgets, escape org scoping, or read another organization's data.

(Note: `compose_email` does send directly — it is not gated behind a draft.)

## Reporting an issue

Found a security problem? Email the maintainers via the
[repository](https://github.com/serenelion/ECN-Monorepo) rather than opening a
public issue with sensitive details.