Docs

Build with Artificial Wit AI

The fastest path from "we should try AI on this" to a working tool call. This page covers what each part of the platform does and how the pieces connect — for hands-on configuration, the in-app screens have their own inline guidance.

Getting started

Three steps to a working tool call

Most teams reach a working, governed tool call in minutes once the first API is configured — no credit card required to start.

1. Configure

Add your APIs, credentials, and knowledge bases inside Artificial Wit AI (ai.artificialwit.com).

2. Authorize

Generate an API key from your user profile.

3. Connect

Point any MCP client at https://api.artificialwit.com/mcp with an x-api-key header set to that key.

MCP server

Artificial Wit exposes a single hosted MCP endpoint. Every API you configure becomes a standardized tool, discoverable by any MCP-compatible client — Claude Desktop, Cursor, ChatGPT, Gemini, n8n, Zapier, or a custom orchestrator — without a separate integration per model.

Endpoint

https://api.artificialwit.com/mcp

Claude Desktop config

{
  "mcpServers": {
    "artificialwit": {
      "url": "https://api.artificialwit.com/mcp",
      "headers": { "x-api-key": "YOUR_API_KEY" }
    }
  }
}

Connect it once, and adding a second or third AI client is configuration, not integration work.

Read the full MCP setup guide

Also an MCP client

Artificial Wit isn't only a server — it's also an MCP client. The Connectors screen auto-detects and speaks both legacy SSE transport (GET /sse + POST /messages, protocol 2024-11-05) and modern Streamable HTTP (POST /mcp, protocol 2025-03-26, mcp-session-id header), so you don't need to know which one a given external MCP server uses. OAuth for third-party connectors — PKCE, RFC 8414 discovery, RFC 7591 dynamic client registration — is handled automatically, so connecting to a server like GitHub's or Notion's doesn't require hand-configuring an OAuth app.

That means every MCP server your team already uses can sit behind the same governed roof as the tools you host yourself.

Knowledge base (RAG)

Ingest PDFs, Word documents, text files, and URLs into a knowledge base. Each document tracks real-time embedding status — Pending, Embedding, Ready, Failed — so you always know what's queryable.

Every answer a RAG-grounded agent gives back carries a source document, a similarity score, and a reference number, so a reviewer can verify the claim instead of taking the model's word for it. Retrieval respects your existing access controls: an agent only surfaces documents the requesting user is already authorized to see. You can curate and assign multiple knowledge bases per agent.

Talk to an engineer about knowledge base setup

API / Tools

The API/Tools registry is where you define the operations an agent can call — REST or GraphQL endpoints, with pre-built connectors available for SAP, Oracle, Salesforce, and NetSuite so the "expose" step is configuration rather than custom API development.

Each tool config sets:

Method and URL

Including global-variable substitution ({{varName}}, resolved server-side) so the same tool can point at different environments without hardcoding values.

Auth

How the tool authenticates against the underlying API, using centralized, reusable credentials.

Access

Public, Protected, or Private, matching the access model used across the rest of the platform.

A built-in Tool Testing Console— request panel, response panel, environment/variable panel, MCP panel — lets you test a tool's actual request and response before wiring it into an agent, so you're not debugging a broken call for the first time in production.

Agents

Configure agents with their own system prompt, assigned LLM (OpenAI, Anthropic, Azure, Gemini, or a local model), knowledge base assignments, tool assignments, and role-gated access. Agents run as one of four types:

Standard

Single-turn responses.

Sequential

Steps run in order.

Parallel

Steps run concurrently.

Loop

Repeats until a condition is met.

That range means a support agent, a sales copilot, and a multi-step compliance workflow can all run on the same platform, each with its own guardrails and orchestration logic — not one generic chatbot stretched across every use case.

Workflows & artifacts

A tool's response doesn't have to come back as a paragraph of text. The UI — a form, table, or chart — is generated dynamically from the tool's own input argument schema: field names, types, required/optional, enum options. Define a ticket-creation tool with order_id, issue_category, and priority arguments, and a matching form appears automatically, with no separate front-end build.

This shows up in two places: full-page, sidebar-navigable artifact dashboards (with a window.cowork.callMcpTool() bridge for calling tools directly from the dashboard), and inline chat artifacts — charts, tables, and forms that render as part of a single message and can expand into a side-panel preview mid-conversation.

Request a walkthrough

Governance & access control

One access control model spans every resource type on the platform — Knowledge Items, Agents, LLM models, APIs, API Credentials, Global Variables, Users, and Pages. Each resource is either Public (any authenticated user) or Restricted (specific roles and/or users). Learn the model on one screen, and you already know how to use it everywhere else.

Role-based access control per agent, per tool, and per knowledge base
Centralized credentials, not duplicated across integrations
Multi-tenant branding with isolated session and tenant state
Audit logs and dedicated/VPC deployment on the Enterprise tier

Read our security overview

Pre-built apps

Beyond registering a custom API, Artificial Wit ships two pre-built, configurable app types: Asset Management and Ticket Management. Each supports mail and print templates — a ticket-confirmation email or a printable asset label — as part of its own configuration, on top of the same dynamic form/table/chart UI.

Asset Management

Track assets with configurable fields, printable labels, and workflow automation.

Ticket Management

Handle support tickets with email templates and status-driven workflows.

FAQ

Frequently asked questions

Do I need to build my own MCP server?

No. Configure your APIs and knowledge bases inside Artificial Wit, and the hosted endpoint at https://api.artificialwit.com/mcp exposes them as MCP tools automatically.

Which AI clients can connect?

Any MCP-compatible client: Claude Desktop, Cursor, ChatGPT (via Custom GPT Actions), Gemini, n8n, Zapier, or a custom orchestrator.

Can Artificial Wit connect to MCP servers I don't host myself?

Yes. It also acts as an MCP client — the Connectors screen handles transport detection and OAuth for external servers automatically.

How is access controlled?

Every resource — agents, tools, knowledge bases, credentials, and more — is Public or Restricted to specific roles/users, enforced consistently across the platform.

Do you support SSO or SCIM?

Yes. SSO/SAML and SCIM provisioning are available on the Enterprise plan.