Integrations and AI

Connect to any system and any AI service.

Import an API from its OpenAPI document, connect a database, start from a ready-made connector or call any endpoint — once, with encrypted credentials — and use those operations in every workflow, agent, report and MCP client. Then put any AI model to work on the same governed connections.

Integration guide Browse 44 connectors
ready-made connectors
44
ready-made operations
390
AI providers, plus any OpenAI-compatible endpoint
10
REST or OData API, database or webhook
Any

Any REST or OData API

Import the API. Get every operation.

Drop in an OpenAPI 2 or 3 document or a Postman collection: UserTasks previews every operation, creates the integration from its server and security scheme, and one reference per operation with typed inputs.

  • API key, bearer, basic or OAuth 2.0 (authorization code with PKCE, automatic refresh)
  • OData v2 / v4 services with standard query options
  • A reference is a typed, tested operation that workflows, agents, reports and MCP call by name
  • Credentials are never taken from the document — and never returned by the API

OpenAPI and Postman import References OAuth connections

Import API Specification: an OpenAPI 3.0 document for a demo logistics API with its name, base URL, tags and bearer authentication, and eight operations selected to become references
The Insert data picker on a workflow step listing the Get Customers response — total, page and a data list with id, name and email — and the choices for a field inside a list: loop through each item, the first item, an item at an index or one found by a condition

Response models and the data picker

Click the field you need — not a JSON path.

Every reference knows the shape of its response: from the OpenAPI schema, the table columns, the connector’s outputs or a real test call. Workflows browse it in the data picker.

  • Pick response › data[] › email and choose: loop through each item, the first item, or one that matches
  • Example values from the captured sample; credential-like fields masked
  • Add a named output from any field with one click

Response models Using API responses

Connectors

44 connectors, ready when you are.

A connector asks only for that service’s settings and credentials, tests the connection with a service-specific check and creates ready-made references. This list is generated from the product’s connector registry.

CRM, ERP, HR and IT service management

Read and update the systems of record your processes run through.

  • SalesforceCRM operations, SOQL queries, and Bulk API for large data loads12 ready-made operations
  • SAP ODataConnect to SAP S/4HANA, SAP BTP, or SAP Business Suite OData services — full CRUD, batch, and function imports15 ready-made operations
  • WorkdayHCM and Financials — full REST coverage with verified shortcuts for Core HCM, Absence, Time Tracking, Recruiting, Talent & Performance, Compensation, and Payroll, plus WQL, RaaS custom reports, and a universal passthrough for every other Workday endpoint45 ready-made operations
  • UKG ProHR, payroll, time management, compensation, and benefits data from UKG Pro (formerly UltiPro)13 ready-made operations
  • ServiceNowIT service management — manage incidents, requests, changes, users, and query any table via the Table API10 ready-made operations
  • Microsoft DataverseQuery, create, and manage data in Microsoft Dataverse (Power Platform)6 ready-made operations
  • JiraApp management — create, update, search issues, manage sprints and boards in Jira Cloud12 ready-made operations

Databases

Query, insert, update and delete with typed parameters — no SQL pasted into workflows.

  • PostgreSQLConnect to PostgreSQL databases — run queries, insert, update, and manage data10 ready-made operations
  • SQL ServerConnect to Microsoft SQL Server and Azure SQL — run queries and manage data10 ready-made operations
  • MySQLConnect to MySQL and MariaDB — run queries, manage data, and call stored procedures10 ready-made operations
  • Oracle DatabaseConnect to Oracle Database — run queries, manage data, call stored procedures and functions12 ready-made operations
  • SnowflakeConnect to Snowflake — run queries, manage data, and call stored procedures10 ready-made operations
  • MongoDBConnect to MongoDB — query, insert, update, delete documents and run aggregations11 ready-made operations
  • FirebaseConnect to Firebase Firestore with a Google service account to query, create, update, and delete documents6 ready-made operations

Files and storage

Upload, download and move the documents a process produces.

  • AWS S3Upload, download, and manage files in Amazon S3 buckets6 ready-made operations
  • Azure Blob StorageUpload, download, and manage blobs in Azure Storage containers4 ready-made operations
  • Google DriveUpload, download, and manage files in Google Drive11 ready-made operations
  • OneDrive for BusinessUpload, download, share, search, and manage files and folders in OneDrive for Business via Microsoft Graph11 ready-made operations
  • SharePointManage files, lists, and sites in Microsoft SharePoint Online15 ready-made operations
  • BoxCloud content management — upload, download, search, share files, and manage folders in Box13 ready-made operations
  • DropboxUpload, download, search, share files, and manage folders in Dropbox12 ready-made operations
  • SFTPUpload, download, list, move, and manage files on remote servers via SFTP (SSH File Transfer)10 ready-made operations

Email and messaging

Notify people where they already work.

  • GmailSend, read, search, label, and manage emails in Gmail via Google API10 ready-made operations
  • Microsoft Exchange OnlineSend and read emails, manage calendars, contacts, and mailbox folders via Microsoft Graph10 ready-made operations
  • SendGridSend transactional and marketing emails via SendGrid2 ready-made operations
  • SlackSend messages, manage channels, search, upload files, and react to messages in Slack15 ready-made operations
  • Microsoft TeamsSend messages and cards to a Microsoft Teams channel using an Incoming Webhook2 ready-made operations
  • TwilioSend SMS, make voice calls, verify phone numbers, and manage communications8 ready-made operations

Documents, signatures and automation

Signatures, shared documents, task lists and other automation tools.

  • DocuSignSend envelopes for signature, check status, download documents, and manage templates8 ready-made operations
  • Google DocsCreate, read, and edit Google Docs — insert text, replace content, manage formatting6 ready-made operations
  • Google TasksCreate, update, complete, and manage tasks and task lists in Google Tasks10 ready-made operations
  • n8nTrigger workflows, manage executions, and orchestrate automations in n8n10 ready-made operations

Payments and banking

Charge, pay out and reconcile.

  • StripePayments, customers, subscriptions, invoices, and refunds via Stripe16 ready-made operations
  • Mercury BankAccounts, cards, transactions, payments, recipients, statements and treasury via the Mercury API13 ready-made operations

Workday, SAP, Salesforce, ServiceNow

The big systems of record have connectors with their common operations ready — and their REST and OData APIs stay fully reachable as references, so nothing is out of bounds. The gallery has working packages for them.

Connector reference Package gallery

Databases, webhooks and HTTP

Reach data where it lives. Let other systems call in.

Databases

PostgreSQL, SQL Server, MySQL, Oracle, Snowflake and MongoDB as integrations: saved, parameterised operations with response models from the table columns. Self-hosted servers can reach databases on private networks.

Database drivers

Inbound webhooks

A Webhook trigger starts a run as soon as it is queued; a Middleware trigger keeps the request open and answers with the run’s variables, so a workflow can validate or enrich a request in line.

shell
curl -X POST "$USERTASKS_URL/api/webhooks/inbound?hook=invoice-intake" \
  -H "Authorization: Bearer <secret>" -d @invoice.json

Outbound calls

Webhook integrations and HTTP Request steps call other systems. Failed deliveries are retried and dead-lettered with an event; every call is logged with method, host, status and duration, headers redacted.

Webhooks Integration types

AI models and agents

Any AI model — doing real work, under your rules.

An agent is a model, instructions, the tools you grant it and an output schema. Run it as a workflow step or on demand. Nothing it does bypasses permissions, and every run is traced.

Ready-made AI connectors

  • Azure OpenAI
  • Claude (Anthropic)
  • DeepSeek
  • Google Gemini
  • Groq
  • Mistral AI
  • Ollama
  • OpenAI
  • Perplexity AI
  • Together AI
  • Any OpenAI-compatible endpoint

Granted tools

Any integration reference, starting a workflow, tools from MCP servers you register, and built-in tools — only what you grant.

Files

Hand an agent the files from a form or a loop: images and PDFs to models that read them, extracted text from Word, Excel and PowerPoint for the rest.

Structured outputs

Declare the outputs; the agent’s answer is parsed into typed variables the next step can branch on.

A person decides

Follow the agent with a User Task, so a person approves, corrects or rejects before anything irreversible happens.

Runs, steps, tool calls, token counts and errors are recorded and emitted as events; prompts and model outputs are not exported. Rate limits apply per workspace, and models can run on your own hardware with Ollama.

AI agents Agents with a person in the loop Triage package

shell
usertasks agents execute <agent-id> \
  --set message="Classify invoice INV-20931 and check the vendor"
usertasks agents execute <agent-id> --set message="…" --set async:=true

MCP, both ways

AI assistants drive UserTasks. Agents use your MCP servers.

The UserTasks MCP server

Claude, ChatGPT, Cursor and other MCP clients connect to /api/mcp with an API key or MCP OAuth. They list and start workflows, act on tasks, build workflows, decision tables and reports, and read the event stream — within the credential’s permissions, every change audited.

json
{ "mcpServers": { "usertasks": {
  "url": "https://tasks.example.com/api/mcp",
  "headers": { "Authorization": "Bearer <your-api-key>" } } } }

MCP server and tools

Your MCP servers as agent tools

Register external MCP servers under Integrations and grant their tools to an agent next to your references and workflows. The agent sees each tool’s name, description and input schema; the run trace shows every call.

  • Registered once per workspace, granted per agent
  • Mixed freely with references, workflows and built-in tools
  • Permissions and rate limits as for every other tool

Agent tools

For your own code and your platform

Every operation from code — and in your identity and log stack.

REST API, SDK and CLI

Every screen is an endpoint described by OpenAPI 3.1; a typed TypeScript SDK and the usertasks CLI cover all of it.

Config as code and GitHub sync

Export a workspace to files, review it in a pull request, diff and apply it — or let GitHub sync do it. Secrets stay out of Git.

Packages

Bundle a workflow and everything it uses into a signed, versioned file and install it in other environments — or from the gallery. Credentials never travel.

SSO and directory sync

OIDC and SAML sign-in; people, groups and managers from Entra ID, Okta, Google Workspace, LDAP or SCIM.

API keys and OAuth apps

Scoped keys, limited to chosen workflows or tables if you like, and OAuth apps with scopes for other applications.

Log export

Runs, steps, integration calls, agent runs and audit events to Splunk, Datadog, OpenTelemetry or a signed webhook.

ts
import { apiKeyAuth, createClient } from '@usertasks/sdk'

const client = createClient({ baseUrl: 'https://tasks.example.com',
  auth: apiKeyAuth(process.env.USERTASKS_API_KEY!) })

await client.POST('/workflows/start', {
  body: { workflow_id: '<id>', variables: { amount: 1200 }, async: true },
})

Start runs, complete tasks and read results from your own applications, scripts and CI — with the same permissions as every other client.

REST API TypeScript SDK CLI Config as code

Questions

Can UserTasks connect to a system that has no connector?

Yes. Any HTTP/JSON API is a REST integration: import its OpenAPI 2 or 3 document or Postman collection to get one reference per operation, or define references by hand. OData v2/v4 services, SQL and MongoDB databases and outbound webhooks are integration types of their own. Connectors only add a setup form, a connection test and ready-made references on top.

How are credentials kept safe?

Credentials are encrypted at rest and write-only: the API, exports and packages return a placeholder instead of the value. OAuth 2.0 connections use authorization code with PKCE and refresh tokens automatically. Outbound calls to private and cloud-metadata addresses are blocked unless a self-hosted server allows private egress, and every call is logged with headers redacted.

Which AI models can agents use?

Any model behind an OpenAI-compatible chat-completions endpoint with tool calling. Ready-made connectors cover OpenAI, Azure OpenAI, Claude, Google Gemini, Mistral, DeepSeek, Groq, Together, Perplexity and Ollama for models on your own hardware.

Can Claude or ChatGPT drive UserTasks?

Yes. UserTasks runs an MCP server at /api/mcp. Claude, ChatGPT, Cursor and other MCP clients can list and start workflows, act on tasks, build workflows, decision tables and reports and read the event stream — always within the permissions of the API key or OAuth token they use.

Do prompts or responses leave my environment in logs?

No. Agent runs, tool calls, token counts and errors are recorded and emitted as events, but prompts and model outputs are not included in exported events, and file contents never reach logs.

Connect it once. Use it everywhere.

The same references serve workflows, agents, reports, lookups on forms and MCP clients — tested, logged and governed in one place.

Integration guide Complex workflowsIntegration hub use case