Connect an AI client

ApexGenius brings your apps and skills into one connection you can take from agent to agent. Connect your apps and keep reusable instructions as skills in ApexGenius, then authorize each compatible AI agent to use that setup. Each agent needs its own setup and access choices. Skills load when the agent reads them, and each agent keeps its own conversations and settings. Client plans, workspace policies, and supported connection methods can differ.

Choose an authentication method

Method	When to use it	What you provide
OAuth with PKCE	Your agent supports browser authorization for remote MCP servers. Start with its app or plugin setup.	The server URL, then approval in your browser.
Scoped API key	Your agent or custom connector supports a remote MCP connection with a securely stored Bearer header.	A dedicated ApexGenius key with the connections and tools you choose.

Both methods use the same product gateway:

https://server.autom8.pro/mcp

The gateway accepts MCP requests over HTTPS. An API key authenticates the request; the client still needs to support MCP or implement its JSON-RPC calls.

Use https://server.autom8.pro/mcp for your connected apps and skills. The documentation site’s MCP search endpoint searches public documentation and does not connect to your app accounts.

Before you connect

Create or sign in to an ApexGenius account.
Connect the supported apps you want the agent to use. Check the account and any selected resources for each connection.
Add any reusable skills you want to make available to the agent.
For Salesforce work, create the intended project and authorize the correct production org or sandbox.
Confirm your AI client supports your chosen authentication method and that your workspace permits the connection.
Check your ApexGenius account access in Settings → Billing.

Authorizing an AI agent to ApexGenius is separate from authorizing an app such as Google Drive or Salesforce to ApexGenius. Each app keeps its own permissions, credential requirements, and limits. See Connect your apps.

Connect with OAuth

Start with the guide for your client. For Codex, open Plugins in the ChatGPT desktop app; manual MCP settings are an alternative in its guide.

Claude

Add a custom connector and approve ApexGenius.

ChatGPT

Create a custom MCP app where your plan and workspace allow it.

Codex

Start in Plugins in the ChatGPT desktop app. Desktop settings and optional CLI setup are also covered.

Cursor

Add a remote server in Cursor MCP settings.

VS Code

Add an HTTP server in mcp.json.

Add ApexGenius in your AI client

Use its supported app, plugin, connector, or remote MCP setup. If it asks for a server URL, enter https://server.autom8.pro/mcp.

Follow the browser authorization opened by the client. Sign in to the ApexGenius account that holds your apps and skills, then review the requested access before approving.

Return to your agent

Enable ApexGenius in the conversation or task, if your client requires it. The client handles the OAuth token exchange and supported token refresh.

Check the saved access

Open Connected AI in ApexGenius and review the agent’s selected connections and scope before its first task.

OAuth details for connector developers

ApexGenius supports the OAuth authorization-code flow with PKCE S256 and the refresh_token grant. Public clients use token_endpoint_auth_method: none; the PKCE verifier protects the authorization-code exchange.

Purpose	Endpoint
Protected-resource metadata	`https://server.autom8.pro/.well-known/oauth-protected-resource/mcp`
Authorization-server metadata	`https://server.autom8.pro/.well-known/oauth-authorization-server`
Client registration	`https://server.autom8.pro/oauth/register`
Authorization	`https://server.autom8.pro/oauth/authorize`
Token exchange and refresh	`https://server.autom8.pro/oauth/token`

Discover the current metadata before configuring a custom client. Register the client’s redirect URI, generate a verifier and S256 challenge, and request an authorization code for the advertised mcp scope and https://server.autom8.pro/mcp resource. The authorization request must include the S256 challenge; the token exchange must include its matching verifier and the registered redirect URI.Keep the resource consistent through authorization, token exchange, and refresh. Store tokens securely and replace them with the values returned by a successful refresh. Send the access token as Authorization: Bearer YOUR_ACCESS_TOKEN. A user’s existing Google or Salesforce token is not an ApexGenius gateway credential.

Connect with an API key

Use a dedicated key for each agent or custom connector so its access can be managed separately.

Add an agent in ApexGenius

Open Connected AI, choose Add an agent, then select Another AI agent. This opens the manual API-key path.

Choose connections and tools

Keep only the app connections this agent needs. Choose Read only for lookups, Custom for specific read and write tools, or Full access for every enabled tool on the selected connections. For a Custom profile, include the skill tools the agent needs if it should discover and read your reusable instructions.

Create and store the key

Select Create the key. ApexGenius shows the complete key once. Copy it directly into the client’s secure credential field or secret store; it cannot be displayed again later.

Configure the client

Set the MCP URL to https://server.autom8.pro/mcp. Configure an Authorization header with the value Bearer YOUR_APEXGENIUS_API_KEY, using the client’s secure credential reference for the real value.

Authorization: Bearer YOUR_APEXGENIUS_API_KEY
Content-Type: application/json

Use the client’s own configuration format. The gateway reads the Bearer credential from the Authorization header; X-API-Key and URL query parameters are not substitutes. Keep the real key out of conversations, source control, screenshots, and public configuration examples. For a custom API connector, send MCP JSON-RPC requests with HTTPS POST to the same endpoint. This discovery request lists the providers available to the key without fetching their records:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "apex_search_tools",
    "arguments": { "query": "" }
  }
}

Discover the exact operation and input schema before calling it. Follow the MCP tool reference for read and write routing. Discovery alone does not verify a real app read.

Muse and other custom clients: this describes the existing ApexGenius gateway contract. The client must support the request format and secure Bearer authentication. A published guide or an accepted API key does not establish that the full workflow has been tested in that client.

Manage access and replace a key

In Connected AI, open the agent to review its scope and selected connections. Save changes applies the updated access to subsequent calls. Provider permissions and connection-level settings still limit what the agent can do; a broader client scope cannot grant new access at the provider. The current key-creation screen has no expiry setting. Use Revoke access when an agent should stop connecting. Revocation stops that agent’s current key or OAuth grant from working and preserves its Activity history. If a key is lost or exposed, create a replacement through the same scoped setup, put it in the client’s secure credential store, and verify a read. Revoke the old agent’s access when it should no longer work. If exposure requires immediate cutoff, revoke it first and expect that client to stop until the replacement is configured.

Older keys can retain legacy access. Review the actual saved profile before treating an existing key as Read only or Custom. Write-confirmation behavior also varies by client and operation.

Verify apps and skills

Open a new conversation with ApexGenius enabled and ask:

Show the apps available through my ApexGenius connection and find a relevant
skill for the task I describe. Read its instructions before using it.
Make no changes to my apps.

Then choose one connected account and run a small read, such as reading a document you select or listing a few issues in a project you name. Check the returned account, source link, and result against that app. Confirm the corresponding successful operation in Activity. Skills provide guidance when the agent reads them. They do not grant app permissions or run uploaded scripts. If an agent cannot list or read a skill, check its selected skill tools and the skill’s availability to your ApexGenius account.

Before a write, confirm the target account, record or file, expected change, and the confirmation behavior of your client.

Troubleshooting

The browser never opens on the OAuth path. Confirm the client supports OAuth for remote MCP servers and that the URL ends in /mcp. For a client that supports secure Bearer authentication instead, use the API-key path above.
An API-key request returns 401. Check the saved credential, the Authorization: Bearer format, and whether access was revoked or the key expired. Never paste the real key into a troubleshooting conversation.
Authorization completes but tools are missing. Confirm ApexGenius is enabled for the conversation, refresh the client’s tool discovery, and review its saved access in Connected AI.
A tool returns a scope or permission error. Check the selected agent connections and tools, then the underlying app permissions. Use an operation within that access or have the owner review the required change.
A tool cannot find an app. Open ApexGenius Connections and verify the app is connected, healthy, and available to your account.
A Salesforce result comes from the wrong environment. List projects, then select the intended production or sandbox project before continuing.
The route returns an account-access message. Open Settings → Billing to review your account state or contact support.
A custom client gets HTTP 200 but no useful result. Inspect the JSON-RPC error and the tool result’s isError. HTTP status alone does not establish success. Respect Retry-After on rate-limit or temporary-service responses.

For route, metadata, and connection-specific fixes, use Common issues.

Start here