Integrations & Tools — Connect AI Agents to Your Stack
Overview
Langoedge goes beyond simple chat by allowing agents to interact with the real world. You can give your AI agents access to built-in tools, custom APIs, or 3,000+ third-party apps via MCP connectors — all authenticated securely through the Connect page.
Definition — Integration Tool: A node type that enables an AI agent to perform actions outside the conversation — sending emails, querying databases, updating CRMs, or triggering automation workflows.
There are three ways to give an agent a capability, and it's worth knowing which one you need:
| Route | Use it for | Setup |
|---|---|---|
| MCP Connector | Mainstream SaaS apps — Slack, Gmail, Airtable, HubSpot, Notion, Stripe, and 3,000+ more. | Connect the account, add the node, pick tools. No code. |
| Native tools | Web crawling, RAG retrieval, sandboxed code, academic and reference lookup, SMS. | Configure the node. No account needed. |
| API Webhooks | Your own REST endpoints, or a service with no MCP-backed app. | Configure URL, method, headers. |
Connect Page
Link and manage every third-party account on the **Connect** page. An OpenAI connection is required before the platform will let you build anything — see the [MCP Connectors & Connect Page guide](/guides/mcp-connectors).
Native Tools
Langoedge includes several built-in tools that require no external setup.
Web Scraper
Read and convert any live URL into markdown. Perfect for agents that need to stay updated on current events, research companies, or browse external websites.
Knowledge Base (RAG)
Upload PDF, DOCX, or TXT files. Langoedge converts them into vector embeddings for fast similarity search. This is the backbone of FAQ chatbots.
API Webhooks
Connect to any external service with a REST API. Supports GET, POST, PUT, DELETE methods with custom headers and vault-authenticated secrets.
Reference Lookup
Built-in `arxiv_search` and `wikipedia_search` tools for academic papers and encyclopedia facts. For general web search, connect a search app (Tavily, Exa, Google) through an MCP Connector.
App Integrations — MCP Connectors
Third-party apps reach your agent through MCP connectors. Rather than Langoedge shipping a hand-written wrapper per app, a connector discovers an app's tools live at runtime — so Gmail, Slack, Teams, Google Calendar, Airtable, LinkedIn, Google Drive, HubSpot, Notion, Stripe and 3,000+ others are reachable from configuration alone.
The workflow is always the same, whatever the app:
Connect the account
Add an MCP Connector node
Scope and attach
Integrations kept hand-written
A few integrations deliberately bypass MCP because they need domain-specific behaviour or sit on the real-time voice path: Cliniko, ServiceM8, Twilio, Telnyx, Pinecone, and Qdrant. These have their own dedicated tools and configuration.
Security & Authentication
Critical: Your credentials — OAuth tokens, API keys, database passwords — are never exposed to the LLM or the frontend. The model sees a tool's inputs and whatever the app returns, nothing else.
For MCP connectors, credentials are held by the connect layer and referenced by an opaque account id. Langoedge stores the link, not your password. Access tokens are minted fresh per request at the moment a tool is invoked — never baked into a compiled graph, since graphs are compiled once, cached, and invoked arbitrarily later. Every call is scoped to the graph owner's connected accounts, so a graph can never reach another user's linked apps.
For API tools and hand-written integrations, the backend injects the required authentication headers (e.g. x-langoedge-secret) into the HTTP request on behalf of the user, keeping the execution environment isolated.
Header Authentication Priority
When making API tool requests, Langoedge merges headers in this order:
- Standard headers from the user's browser HTTP request.
- Secure account tokens fetched from the credentials vault database.
- Custom headers configured directly in the Graph Node editor.
Performance Best Practices
Use Background Execution
Enable background mode for slow tools (heavy database queries, report generation). The agent triggers the task and continues the conversation while it runs.
Leverage Caching
Langoedge caches compiled graphs in RAM. Edit frequency and timestamp changes invalidate the cache automatically.
Stream Responses
Use Server-Sent Events (SSE) for real-time token streaming instead of waiting for full responses.
Minimize Tool Nodes
Fewer tool calls per turn means lower latency. Combine related operations into a single tool when possible.