Connect your coding assistant to Veryfront with OAuth. Use your Veryfront account to approve access. The client stores and refreshes its credentials. You need a client with Streamable HTTP and OAuth support.
Claude Code
-
Add the server:
bash claude mcp add --transport http veryfront https://api.veryfront.com/mcp -
Start Claude Code and run
/mcp. Select veryfront, authenticate, then sign in to Veryfront and select Approve. Check the app name, signed-in account, and return host before approving. -
Ask: Use Veryfront to list my projects. The client calls
veryfront__list_projectsand returns the projects you can access.
See Claude Code MCP setup for client configuration options.
Codex
-
Add the server:
bash codex mcp add veryfront --url https://api.veryfront.com/mcp -
Authenticate, then sign in to Veryfront and select Approve:
bash codex mcp login veryfront -
Open Codex and ask: Use Veryfront to list my projects. Use
/mcpto inspect the connection and available tools.
See Codex MCP setup for client configuration options.
Cursor
-
Add the server to your project's
.cursor/mcp.json, or merge this entry into an existingmcpServersobject:json { "mcpServers": { "veryfront": { "url": "https://api.veryfront.com/mcp" } } } -
In Cursor's MCP settings, authenticate veryfront. Sign in to Veryfront and select Approve.
-
In Agent chat, ask: Use Veryfront to list my projects. Approve the tool call when Cursor prompts you.
See Cursor MCP setup for global configuration and client settings.
Check access
The clean /mcp URL lists canonical names such as veryfront__list_projects. No naming query parameter is required. An empty project list is valid if your account has no projects.
The connection uses your existing Veryfront permissions, including authorized writes. OAuth does not grant access to another user's resources or connect external services for you. Connect an integration before calling tools that need an external account.
For a project connection, use https://api.veryfront.com/projects/{project_reference}/mcp, replacing {project_reference} with the project UUID or URL-encoded slug. The endpoint resolves that project and checks your access.
If authentication fails, check the server URL and sign in again. If a tool call fails, inspect its permissions and arguments in the MCP reference.
Revoke access
In Veryfront, open Settings → Authorized apps. Select Revoke beside the client and confirm. The client's next MCP request fails. Connect and approve again to restore access.
Revoking an app invalidates its tokens. API keys, memberships and invitations created through the app remain. Review and remove those separately if they are no longer needed.
Removing a server from a client's local configuration does not revoke its Veryfront grant. Revoke the grant in Veryfront when you stop using a client.
Alternative: API key
For CI or headless clients, create an API key and use the same server URL with the header Authorization: Bearer <API_KEY>. Existing API-key connections remain supported.
For Codex, read the key from an environment variable set by your secret manager:
codex mcp add veryfront-headless --url https://api.veryfront.com/mcp --bearer-token-env-var VERYFRONT_API_KEYFor OpenCode, read the same environment variable from opencode.json:
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"veryfront": {
"type": "remote",
"url": "https://api.veryfront.com/mcp",
"enabled": true,
"headers": {
"Authorization": "Bearer {env:VERYFRONT_API_KEY}"
}
}
}
}Run opencode mcp list to inspect the connection.
For other clients, configure the Authorization header through their supported secret settings. Keep API keys out of shared configuration files and source control. API-key access is managed through the key's permissions and lifecycle, separately from Authorized apps.
Debug a run
In Studio, open a failed agent run and select Copy debugging prompt. Paste the prompt into your connected coding agent. The prompt names the project and the run, and asks the agent to:
- Read the run with
get_run. - Read its events with
list_run_events. - Read its child runs with
list_run_child_runs. - Explain what failed and propose a fix, for example with
update_agent.
Review the proposed change before the agent applies it, then run the agent again to confirm the fix.
Related
- Tool names and compatibility: Preserve an existing client's discovery names.
- List projects: Inspect the first read-only tool call.