MCP Server

OutCallerAI exposes an MCP server that lets AI coding agents manage your workspace - agents, usecases, leads, calls, and exports - all through natural language.

How it works

The MCP server exposes 23 read and write tools across six resources. AI agents can manage workspace settings, usecases, voice agents, leads, calls, and export jobs. All tools require an API key and respect workspace permissions.

Two transports are available:

Transport	Best for	How it runs
stdio	Local AI clients (Cursor, VS Code, Claude Desktop)	Runs as a child process on your machine
hosted	Cloud AI agents (Codex, Gemini CLI)	Connects to `https://api.outcallerai.com/mcp`

Client setup

Set your API key, then add via CLI:

export OUTCALLERAI_API_KEY=your-api-key
codex mcp add outcallerai \
  --url https://api.outcallerai.com/mcp \
  --bearer-token-env-var OUTCALLERAI_API_KEY

Or in ~/.codex/config.toml:

[mcp_servers.outcallerai]
url = "https://api.outcallerai.com/mcp"
bearerTokenEnvVar = "OUTCALLERAI_API_KEY"

Verify:

codex mcp list

The API key must be set in your shell before starting Codex.

Set your API key, then add via CLI:

export OUTCALLERAI_API_KEY=your-api-key
claude mcp add --transport http outcallerai https://api.outcallerai.com/mcp

Or use stdio via .mcp.json. The server reads OUTCALLERAI_API_KEY from your shell.

{
	"mcpServers": {
		"outcallerai": {
			"command": "npx",
			"args": ["-y", "@outcallerai/mcp-stdio"],
			"env": {
				"OUTCALLERAI_API_KEY": "${OUTCALLERAI_API_KEY}"
			}
		}
	}
}

Verify:

claude mcp list

Create .cursor/mcp.json in your project root. The server reads OUTCALLERAI_API_KEY from your shell environment.

{
	"mcpServers": {
		"outcallerai": {
			"command": "npx",
			"args": ["-y", "@outcallerai/mcp-stdio"],
			"env": {
				"OUTCALLERAI_API_KEY": "${OUTCALLERAI_API_KEY}"
			}
		}
	}
}

Restart Cursor. Ask its agent about your OutCallerAI workspace to verify.

Create .vscode/mcp.json. VS Code prompts for your API key via a password input:

{
	"servers": {
		"outcallerai": {
			"command": "npx",
			"args": ["-y", "@outcallerai/mcp-stdio"],
			"env": {
				"OUTCALLERAI_API_KEY": "${input:outcallerai-api-key}"
			}
		}
	}
}

Run MCP: List Servers to verify.

Add to your Gemini CLI config. The server reads OUTCALLERAI_API_KEY from your shell.

{
	"mcp": {
		"servers": {
			"outcallerai": {
				"command": "npx",
				"args": ["-y", "@outcallerai/mcp-stdio"],
				"env": {
					"OUTCALLERAI_API_KEY": "${OUTCALLERAI_API_KEY}"
				}
			}
		}
	}
}

Add to opencode.json. The server reads OUTCALLERAI_API_KEY from your shell.

{
	"mcpServers": {
		"outcallerai": {
			"command": "npx",
			"args": ["-y", "@outcallerai/mcp-stdio"],
			"env": {
				"OUTCALLERAI_API_KEY": "{env:OUTCALLERAI_API_KEY}"
			}
		}
	}
}

Persist the API key

Running export OUTCALLERAI_API_KEY="your-api-key" sets the key only for the current terminal session. A new terminal, or an application started outside that terminal, will not inherit it.

For zsh, add the export to ~/.zshrc and reload the shell:

# ~/.zshrc
export OUTCALLERAI_API_KEY="your-api-key"

source ~/.zshrc

Use ~/.bashrc for interactive Bash shells or the equivalent startup file for your shell. Start or restart the MCP client after loading the variable.

Shell startup files store values as plaintext. On a shared machine, load the key from a password manager or operating-system keychain instead. For example, on macOS you can store the key in Keychain once:

read -s "OUTCALLER_KEY?OutCallerAI API key: "
echo
security add-generic-password -U \
  -a "$USER" \
  -s OUTCALLERAI_API_KEY \
  -w "$OUTCALLER_KEY"
unset OUTCALLER_KEY

Then add this loader to ~/.zshrc:

export OUTCALLERAI_API_KEY="$(security find-generic-password \
  -a "$USER" \
  -s OUTCALLERAI_API_KEY \
  -w 2>/dev/null)"

Verify that the variable is available without printing the secret:

[[ -n "$OUTCALLERAI_API_KEY" ]] && echo "OutCallerAI API key loaded"

Never commit an API key to a repository or place it directly in a shared MCP configuration file.

Configuration

Variable	Default	Description
`OUTCALLERAI_API_KEY`	(required)	Workspace API key
`OUTCALLERAI_BASE_URL`	`https://api.outcallerai.com`	API origin
`OUTCALLERAI_TIMEOUT_MS`	`30000`	Request timeout

To run the server locally:

export OUTCALLERAI_API_KEY=your-api-key
npx -y @outcallerai/mcp-stdio

Available tools

All tools are grouped by resource. Read-only tools are safe to call; destructive tools permanently delete or cancel data.

Workspace

Tool	What it does
`outcallerai_get_workspace`	Get workspace details
`outcallerai_update_workspace`	Update workspace name

Usecases

Tool	What it does
`outcallerai_list_usecases`	List usecases
`outcallerai_get_usecase`	Get usecase details
`outcallerai_create_usecase`	Create a usecase
`outcallerai_update_usecase`	Update a usecase
`outcallerai_delete_usecase`	Delete a usecase

Agents

Tool	What it does
`outcallerai_list_agents`	List voice agents
`outcallerai_get_agent`	Get agent details
`outcallerai_create_agent`	Create a voice agent
`outcallerai_update_agent`	Update a voice agent
`outcallerai_delete_agent`	Delete a voice agent

Leads

Tool	What it does
`outcallerai_list_leads`	List leads
`outcallerai_get_lead`	Get lead details
`outcallerai_create_lead`	Create one or more leads
`outcallerai_update_lead`	Update a lead
`outcallerai_delete_lead`	Delete a lead

Calls

Tool	What it does
`outcallerai_list_calls`	List calls
`outcallerai_get_call`	Get call details
`outcallerai_create_call`	Initiate outbound calls

Jobs

Tool	What it does
`outcallerai_export_calls`	Export calls as JSON
`outcallerai_get_job`	Check export job status
`outcallerai_cancel_job`	Cancel a running export job

Tool annotations

Every tool includes MCP annotations that tell the client whether the tool is read-only, destructive, or idempotent. This helps AI agents ask for confirmation before destructive actions.

Troubleshooting

Symptom	Likely cause
No tools listed	Confirm Node.js 20 or 22; run the command manually and check stderr
Authentication failure	`OUTCALLERAI_API_KEY` is not set in the child process environment
Hosted 401	The workspace uses OAuth-mode authentication - API keys cannot authenticate. Switch to credential-mode
Timeout	Increase `OUTCALLERAI_TIMEOUT_MS` or check API health
Wrong tools after upgrade	Restart the server and reset the client's MCP tool cache

How is this guide?

On this page