Quickstart

Sign in at https://agentcall.io. Use the same account on your phone and when you authorize your agent’s connection. That account connects your agents to your devices.

You’ll need a client that supports remote MCP servers. The connection steps depend on your client; device setup, the test conversation, and calling rules are shared by all three paths.

1. Register the server — once per machine:

   claude mcp add --transport http --scope user agentcall https://agentcall.io/api/mcp

   --scope user makes the server available to every project on your machine. Without it, the entry is bound to the directory where you ran the command; a session opened elsewhere will not see the tools.

2. Authenticate. Start a Claude Code session, run /mcp, select agentcall and choose Authenticate. In the browser, sign in with the same account you use in AgentCall. Claude Code stores and refreshes the OAuth tokens for you.

Continue to device setup

Add the server to ~/.codex/config.toml. Keep the 300-second timeout: a ring normally waits up to three minutes and stand_by for four, so a shorter client timeout can interrupt the wait.

[mcp_servers.agentcall]
url = "https://agentcall.io/api/mcp"
tool_timeout_sec = 300

Then authenticate once and follow the browser sign-in with the same account:

codex mcp login agentcall

Codex support is in beta. The tool contracts are client-agnostic, but the conversation flow is tested primarily with Claude Code. Identified Codex calls appear in History and on the Agents page as codex.

Continue to device setup

AgentCall works with compatible remote MCP clients for research, planning, operations, coding, or other work. Other clients have not been verified yet; check these requirements before connecting.

• Endpoint: https://agentcall.io/api/mcp, using Streamable HTTP transport.
• Authorization: OAuth 2.1 with PKCE through the MCP authorization flow. Discovery is published at /.well-known/oauth-protected-resource/api/mcp and /.well-known/oauth-authorization-server. Authorize the same account you use in the app. If your client cannot complete OAuth, it can send a fixed Authorization: Bearer header with a token from Tokens for clients without OAuth in Settings.
• Timeouts: allow at least 300 seconds per tool call. When the client requests progress with _meta.progressToken, the server can send notifications/progress every 10 seconds while waiting. Whether that resets the timeout depends on the client; keep the longer timeout even if it supports progress.
• Identification: History and Agents use available client metadata, with a meaningful User-Agent as a fallback. Unrecognized clients may appear as unknown. Supply a work or project name with project on tools that support it; see Tools for the exact arguments.

Continue to device setup

On your phone, open guided setup and sign in with the same account. In Register a device, choose Enable notifications. You can register a desktop browser too. For an existing setup, use Devices & notifications → Ring this device.

iPhone: in Safari, tap Share → Add to Home Screen. Open AgentCall from the Home Screen, then enable notifications inside the installed app. Web push requires that installed app; a lock-screen banner opens it so you can answer.

Android: Chrome supports the web flow directly. Notifications can offer Open and Dismiss actions; Open brings up the call screen, where you answer. An Android app with a real lock-screen ring — Answer and Decline on the lock screen — is in private beta and not yet available to install.

A call requests notifications on your enrolled devices, and the first successful answer joins the conversation. Delivery depends on device permissions, connectivity and platform settings — check the phone you intend to use with a test.

Continue to a test conversation

While you can still check your setup, ask your connected agent:

Call me with AgentCall now so we can test the line.

1. Look for the notification on the phone you want to reach. Open the call screen and answer.
2. Listen to the agent, reply, and check that it responds to what you said.
3. After the call, read the exchange in History.

If the phone stays silent, check It didn’t ring, device permissions, and calling availability. You can also open the call screen to check the current call. Agent-initiated tests follow your normal quiet hours and Do not disturb setting.

Choose when agents call

Ask for the updates and conversations that help you stay involved. Save standing instructions in Your call rules: choose Add rules or Edit rules, write them in Call rules, then choose Save rules. You can also tell an agent what you want for the task at hand:

# follow research as it develops
Research these options. While you're working, call me every hour
with a short status update.

# hear about a milestone
When the report is ready, call me and walk me through the findings.

# make room for your preference
Work autonomously. If a choice depends on my preference,
call me with AgentCall so we can talk it through.

# get a wrap-up call
When you're done, call me and ask what's next.

# stay reachable while you step away
I'm stepping away — stand by for calls.

Agents receive saved rules when they connect. A running compatible agent must follow through on them; timed updates depend on that agent’s scheduling. Saving a rule does not create a recurring schedule in AgentCall. For a call back while an agent is working or standing by, see Callbacks.

• Tools — exactly what each tool does and returns.
• Callbacks — asking an agent to call you.
• Settings — voice, rules, language, devices.