Configuration Reference¶

Choose the instructions for your role: developers connect local IDE to the centralized MCP gateway; organization administrators make it available in business chatbots such as Claude and ChatGPT; deployment operators configure the Registry runtime environment.

• For developers: local client setup
• For organization administrators: business chatbot integration
• For deployment operators: Registry runtime environment

For Developers: Local Client Setup¶

Connect Claude Code, OpenAI Codex, GitHub Copilot, or Cursor on your workstation to an existing Jarvis Registry gateway. Your administrator provides the gateway and grants access; you configure your client and sign in through your organization's identity provider.

Choose Your Gateway URL¶

Sign in to your organization's Jarvis Registry and open Integration guide in the header. Copy the gateway URL from the generated configuration. This preserves your deployment's hostname and base path.

Before connecting, make sure:

• Your workstation can reach the gateway, including any required VPN, private DNS, and trusted TLS certificate chain.
• Your administrator has configured the identity provider and granted your account the necessary roles and resource access.
• Your client is current and supports remote MCP over Streamable HTTP and browser-based OAuth.

Claude Code¶

Add the gateway for your user account, available across projects:

claude mcp add --transport http --scope user jarvis-registry https://jarvis.example.com/gateway/proxy/mcpgw/mcp

Start Claude Code and run /mcp. Select jarvis-registry and complete authentication in your browser when prompted, using your organization account. Run claude mcp list to inspect configured connections.

For a configuration shared through a project repository, run the command from the project root with --scope project instead of --scope user. This writes a .mcp.json entry:

{
   "mcpServers": {
      "jarvis-registry": {
         "type": "http",
         "url": "https://jarvis.example.com/gateway/proxy/mcpgw/mcp"
      }
   }
}

Review and approve project-scoped MCP servers when Claude Code prompts. User-scoped settings are stored in ~/.claude.json; use the CLI to manage them rather than replacing that file with the project JSON above.

See the Claude Code MCP reference for scopes, authentication, and managed configuration.

OpenAI Codex¶

Add this table to ~/.codex/config.toml for your user account. Merge it with existing settings rather than replacing the file:

[mcp_servers.jarvis-registry]
url = "https://jarvis.example.com/gateway/proxy/mcpgw/mcp"

Authenticate and inspect the connection:

codex mcp login jarvis-registry
codex mcp list

Complete the browser sign-in with your organization account. In an interactive Codex session, use /mcp to inspect active servers.

For a shared project configuration, put the same table in .codex/config.toml instead. Codex loads project settings only for trusted projects. Each developer still authenticates independently. These instructions cover local Codex clients; they do not configure an organization connector in ChatGPT.

See the OpenAI Codex MCP reference for supported clients and configuration options.

GitHub Copilot (VS Code)¶

Merge the following entry into .vscode/mcp.json in your workspace root. VS Code uses the servers key rather than mcpServers:

{
   "servers": {
      "jarvis-registry": {
         "type": "http",
         "url": "https://jarvis.example.com/gateway/proxy/mcpgw/mcp"
      }
   }
}

For a connection available across workspaces, run MCP: Add Server from the Command Palette, choose HTTP, enter your gateway URL, and select your user profile as the configuration location instead of the workspace.

Run MCP: List Servers, select jarvis-registry, and start the server. Review any trust prompt and complete the OAuth sign-in in your browser using your organization account. Open Copilot Chat in Agent mode and use the tools picker to confirm that the gateway tools are enabled. Your organization's Copilot policies must permit MCP server use.

See the VS Code MCP server guide for configuration scope, authentication, and tool approvals.

Cursor¶

Merge the following entry into ~/.cursor/mcp.json to make the gateway available across projects, or .cursor/mcp.json in a project root to share it with that project:

{
   "mcpServers": {
      "jarvis-registry": {
         "url": "https://jarvis.example.com/gateway/proxy/mcpgw/mcp"
      }
   }
}

Open Cursor's MCP settings, enable jarvis-registry if needed, and complete the OAuth sign-in when prompted. If the connection does not refresh after editing the file, restart Cursor. Confirm that the gateway tools appear in the client before using them in Agent chat.

See the Cursor MCP reference for configuration locations, authentication, and team distribution.

Verify the Connection¶

1. Confirm that jarvis-registry is connected and authenticated in the client's MCP status view.
2. Ask the assistant to discover an approved tool, for example: "Use Jarvis Registry to find the tools available for our issue tracker." The gateway exposes discovery and execution tools; connecting it does not install every downstream tool as a separate client entry.
3. Invoke a known read-only tool that your account is allowed to use. Check both discovery and execution, not just the connection indicator.

Connecting to the gateway does not grant access to every registered server. Discovery and execution remain subject to the caller's permissions. Some downstream services also require a separate authorization step.

For Organization Administrators: Business Chatbot Integration¶

Make Jarvis Registry available inside your organization's Claude or ChatGPT workspace so business users can access approved tools from chat. Administrators register and approve the connection in the AI platform; users select it and sign in when prompted. Users do not edit JSON/TOML files, run CLI commands, or configure individual downstream MCP servers.

Before You Begin¶

• Have an organization owner/admin account and a plan that supports custom remote MCP integrations. Availability, role names, and menu labels depend on the platform and plan; consult the vendor guides linked below.
• Complete Production Deployment, configure your identity provider, and register the approved MCP servers. Assign the intended users and groups the required Registry permissions.
• Copy the shared gateway URL from Integration guide, for example https://jarvis.example.com/gateway/proxy/mcpgw/mcp. Use the gateway URL, not a user-specific direct server URL.
• Confirm that the AI platform can reach the gateway and its advertised OAuth discovery, registration, authorization, and token endpoints. Use HTTPS with a trusted certificate. A successful connection from an employee workstation does not establish cloud connectivity.

Claude's remote connectors connect from Anthropic's cloud, including when used in Claude Desktop. Your gateway must be reachable from that infrastructure; a private VPN or localhost URL is not sufficient. ChatGPT also requires a remotely reachable endpoint, or a supported Secure MCP Tunnel configured by your administrators. Do not expose an unauthenticated gateway to satisfy network requirements.

Claude: Organization Custom Connector¶

Sign in as an organization Owner or another role permitted to manage connectors.

1. Open Organization settings > Connectors, then select Add and the custom connector option.
2. In Add custom connector, enter Jarvis Registry in Name.
3. Paste your shared gateway URL into MCP server URL, for example https://jarvis.example.com/gateway/proxy/mcpgw/mcp, then select Continue.
4. Review the detected OAuth settings and complete the sign-in setup for your deployment. Keep per-user OAuth authentication enabled rather than selecting No sign in or supplying a shared user's token.
5. Finish adding the connector and review its availability and tool controls for your organization.

For business users: In a chat, select + > Connectors > Jarvis Registry. Sign in with your organization account if prompted. The organization connector can be used in Claude Desktop without editing local configuration files.

See Anthropic's custom connectors using remote MCP for current organization roles, authentication options, and client availability.

ChatGPT: Custom MCP Server / Plugin¶

Sign in to the intended ChatGPT workspace as an admin or owner. Enterprise/Edu may also authorize members to create and test connections; workspace publishing remains an admin/owner action.

1. Open Plugins, select Add, then Add custom MCP server.
2. Enter Jarvis Registry in Name. The icon and description are optional.
3. Under Connection, keep Server URL selected and paste your shared HTTPS gateway URL. Set Authentication to OAuth. Use Advanced OAuth settings only if your deployment requires specific client settings or scopes. If your administrators have configured a Secure MCP Tunnel instead, select Tunnel.
4. Review the risk warning, select I understand and want to continue, and choose Create as a plugin. Complete OAuth sign-in when prompted.
5. Review the discovered tools and use your workspace's admin controls to approve the tools intended for users.
6. Select Publish to workspace (or the equivalent workspace publishing control) and confirm the intended audience. Users can connect only after the integration is published; creating the connection and approving tools alone does not establish organization-wide access.

For business users: In a new conversation, type @ and find and select the approved Jarvis Registry plugin or custom MCP server. Sign in with your organization account if prompted, then try a permitted read-only task.

Use ChatGPT web for administration. To validate publishing, open a new conversation, type @, select Jarvis Registry, and try a permitted read-only task. Confirm support in your deployed ChatGPT desktop app before directing users there. Older workspaces may show different Apps or publishing controls; use OpenAI's guide below for those variants.

See OpenAI's Connecting custom MCP servers in ChatGPT for current creation, publishing, network, and plan requirements.

Business User Access and Validation¶

Admin approval makes the integration available; it does not grant every user access to every Registry server. Users still authenticate individually, Registry roles and resource ACLs still apply, and some downstream services require additional user consent. Platform tool approval governs the gateway tools exposed to the chatbot; it does not replace Registry authorization for downstream servers.

Before enabling the integration broadly, test with both an allowed and a restricted account using the two-user access walkthrough. Verify discovery and a read-only invocation from the actual chatbot, including any downstream authorization prompt. Review write-capable tools before approving them. When gateway tool definitions change, review and refresh the platform's tool list as required; ChatGPT retains an approved snapshot rather than automatically accepting every change.

Do not share an administrator's OAuth session, access token, or refresh token with business users. Keep per-user sign-in enabled, and use platform access controls together with Registry permissions. Jarvis Registry governs requests routed through its gateway, not other integrations users connect outside it.

Gateway Versus Direct Server Connection¶

The header's Integration guide connects to the Registry gateway and is the recommended starting point for access to the shared catalog. A server card's MCP configuration instead connects through the direct proxy to one registered server, using a URL shaped like /gateway/proxy/server/{userId}/{serverPath}. Copy that URL from the interface when a single-server connection is required; do not distribute a user-specific direct URL as the organization's shared gateway URL.

Troubleshooting¶

Registry Runtime Environment Configuration¶

This section is for operators deploying and running Jarvis Registry. These environment files, identity provider settings, authorization rules, and Docker Compose definitions configure the Registry services themselves. They are separate from the MCP client connection files used by developers IDE and the chatbot integration settings managed by organization administrators.

Active Configuration Files¶

Environment Setup¶

1. Create your main environment file:

cp .env.example .env

1. (Optional) Create a local Docker override:

cp docker-compose.override.yml.example docker-compose.override.yml

1. Edit only what you need in .env and docker-compose.override.yml.

Notes¶

• OAuth credential and AgentCore-specific setup content was removed from this page because those flows are now handled in the interface.
• oauth2_providers.yml is the IdP configuration source for authentication providers.
• oauth_providers.yaml is intentionally not included here.