Integrations
Integrations connect agents to external tools using the Model Context Protocol (MCP). An integration exposes a set of tools — typed functions the agent can call at runtime to read data, take actions, or trigger processes in external systems.
The agent decides when to call a tool based on the conversation. You don’t write any routing logic — the model handles it.
Supported integrations
The following integrations are available out of the box. Click Add integration in the dashboard to browse and connect them:
| Integration | Category | What the agent can do |
|---|---|---|
| Slack | Communication | Post messages to channels, read channel history, search conversations |
| GitHub | Developer | List repos, read and create issues, comment on pull requests, read file contents |
| PostgreSQL | Database | Run read queries against a Postgres database |
| Brave Search | Search | Run web searches and get back ranked results with snippets |
| Zapier | Automation | Trigger any Zapier workflow via a natural-language request |
Custom MCP servers
Any MCP-compatible server can be connected. Go to Integrations → Add integration → Custom MCP server and configure:
| Field | Description |
|---|---|
| Name | A label used in the UI and referenced in agent/step configs |
| Transport | stdio for local process-based servers, HTTP for remote SSE endpoints |
| Command | (stdio only) The shell command to start the server — e.g. npx @modelcontextprotocol/server-github |
| Args | (stdio only) Space-separated arguments passed to the command |
| Env | (stdio only) JSON object of environment variables — where credentials live |
| URL | (HTTP only) The SSE endpoint URL |
| Headers | (HTTP only) JSON object of HTTP headers, e.g. {"Authorization": "Bearer ..."} |
stdio example
Running a local MCP server via npx:
{
"transport": "stdio",
"command": "npx @modelcontextprotocol/server-github",
"env": {
"GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..."
}
}HTTP example
Connecting to a remote SSE MCP endpoint:
{
"transport": "http",
"url": "https://mcp.example.com/sse",
"headers": {
"Authorization": "Bearer my-api-key"
}
}Browsing available tools
Once connected, click any integration row in the dashboard to expand it and see all the tools the server exposes. Each tool shows its name, description, and input parameters. Hover over a parameter chip to see its type and whether it’s required.
Attaching to agents
Open an agent → Capabilities → add integrations. The agent will see and be able to call all tools from every attached integration on every turn.
Attaching to workflow steps
Integrations can also be attached at the workflow step level. A step only has access to the integrations explicitly attached to it — not to all integrations attached to the underlying agent. This lets you restrict which tools are available at each stage of a pipeline.
Security
- Credentials are stored encrypted at rest
- MCP server processes are isolated per integration instance
- Credentials are never included in the agent’s context or exposed in logs — only tool results are returned to the model