Connect your agents (n8n, Claude, ChatGPT, code)
The steps every platform shares, then which guide to follow for n8n, n8n Cloud, Claude, ChatGPT and custom code.
Every platform connects to Veefore the same way: an agent with its own identity, a tool that does the action, a policy that decides, and a runtime that tells Veefore where the agent runs. This page does the shared part once. Then follow the guide for your platform.
Connected is not protected
Veefore controls the actions an agent sends through Veefore. A Gmail node wired straight into an n8n workflow, or a Gmail connector added directly in Claude or ChatGPT, never reaches Veefore. The Integrations page says so for each runtime: only a Veefore-managed n8n can be Enforced within scope; the rest are Partial at best.
#Before you start
You need a Veefore account at console.veefore.com
with the Developer role or higher, and about fifteen minutes. All of the
addresses below use the hosted gateway, https://gateway.veefore.com.
Pick the environment
At the top of the console, choose the environment you are working in, for example Development. Everything you create below belongs to it.
Create one agent per platform
Agents → New agent. Name it after where it runs ("Support n8n",
"Claude Code — ops", "Sales GPT"), grant the capability its actions need, for
example email.send, and create it.
Copy the private key and key ID now. They are shown once. Store them in your password manager. One agent per platform keeps each one's status and audit trail separate.
Register the tool
Tools → New tool. For an action Veefore performs itself (the strongest option), choose HTTP: the endpoint URL, the method, the destination allowlist, and a credential.
Add the credential first under Credentials → New (API key or bearer token). The secret is sealed in Veefore's broker; the agent and the platform never see it. See Register a tool and Broker a credential.
Authorize the agent on the tool
On the tool's page, Agents → Authorize, and pick the agent. Holding the capability is not enough; the tool must be authorized too.
Write and activate a policy
Policies → New policy, for example "Email: block outside domains, approve anything marked URGENT, allow the rest". Bind it to the workspace, add a test case, and Activate (this needs a step-up). See Write your first policy.
Turn on enforcement
A new environment starts in Observe, which records but blocks nothing.
When you are ready, switch it to Enforce from the environment switcher or
with veefore deploy --mode ENFORCE. See
Enforcement modes.
Register the runtime
Integrations → Add on the card for your platform, and choose the agent. This is what lets Veefore show whether the agent is connected and how far its actions are enforced. The runtime starts Pending and Unverified.
#Which guide next
| Where the agent runs | How it connects | Guide |
|---|---|---|
| n8n you host yourself | The Veefore node, plus workflow verification | n8n (self-hosted) |
| n8n Cloud | n8n's MCP Client Tool, until the Veefore node is verified | n8n Cloud |
| n8n run by Veefore's profile | Hardened deployment that fails closed | Veefore-managed n8n |
| Claude Code, Claude Desktop, Cursor, VS Code | Veefore's MCP endpoint with an MCP key | Claude and MCP clients |
| ChatGPT custom GPTs, other OpenAPI platforms | The Veefore relay you run | ChatGPT |
| Your own code | The JavaScript or Python SDK | Protect an agent |
| Anything else | The veefore.action/v1 envelope |
Other platforms |
#Check it worked
For every platform, the proof is the same three places:
- Actions lists the call with its decision. A refused call shows the rule that refused it.
- Integrations → your runtime shows Connected once Veefore has seen the agent, and its enforcement status with the reason.
- Audit has the event, with the runtime and, for n8n, the workflow and execution ids.