Aeion Forge + AI Coding Tools
Aeion Forge exposes a single MCP server (JSON-RPC 2.0 + REST over HTTPS, with SSE streaming) — there's no separate integration built per tool. Claude Code has native, first-class MCP client support, so it's the most seamless way to connect. Any other MCP-compatible agent (Cursor, custom runtimes, bespoke agent harnesses) can add the same server as a remote MCP endpoint. Tools that don't yet speak MCP can call Forge's REST/JSON-RPC API directly with an API key. The 160+ Forge actions become available in your AI tool's tool palette either way.
What Aeion Actually Ships vs. What You Wire Up
`
Claude Code Setup Walkthrough
```bash
Cursor Integration via HTTP MCP
```typescript
Aider Integration via REST API
```bash
Devin, and Other Webhook-Capable Agents — a DIY Pattern
```typescript
FAQ
No — there's one MCP server. Claude Code connects to it natively. Any other MCP-aware tool connects the same way. Tools without MCP support call the same REST/JSON-RPC endpoint directly.
A Forge API key with a coarse scope set (`read`, `write`, `admin`, `webhooks`), an environment (live/test), and an optional expiration. There's no per-action allowlisting on a key today — scope is at that granularity, not per-tool-action.
Forge has a real file-lock/contention system: an agent can check for contention on a file before editing, and `attach_commit` can auto-claim locks (policy: skip/claim/force) so parallel agents can see who's working where. It informs rather than hard-blocks by default.
Yes — all 160+ actions are exposed the same way regardless of which client connects; there's no tool-specific subset.
Not required, but recommended for auditability — a named key per tool/developer makes it easy to see which agent made which change and to revoke one without affecting others.
Yes — every API key is scoped to a single tenant, and all Forge data access goes through the platform's normal tenant isolation.