How it worksFeaturesDocsGitHubDashboard

Create a MCP server from any data source in 3 clicks

Arthur MCP is an open-source project built to help any kind of data source connect easily with AI systems. So you can turn your data into smart tools and resources without needing to be a tech expert.

  • Free to use, modify, and self-host
  • Easy integration with Sentry and Prometheus (Metrics and Error Tracking)
  • Docker & Render-ready to deploy in minutes
  • Contributions welcome: issues, PRs, templates
+100REST API templates
18Data source types
4Built-in roles
100%Open source
Terminal

Why Arthur MCP

The hard part isn't the tool schema

AI clients are only useful when they can safely act on real systems. Arthur MCP is built for the operational layer around a tool call: auth, secrets, rate limits, logs, prompts, docs, and a workflow that lets humans review exactly what an AI client can access.

Credentials stay server-side

Secrets, upstream auth, and API keys never render in list views. Reference a secret by name instead of pasting raw values into a server.

Every action leaves a trail

Audit logs record who changed what across servers, tools, secrets, and roles. Error tracking forwards failures to the provider you already use.

Humans review what AI can touch

Four built-in roles and granular per-domain permissions gate every create, edit, delete, test, and reveal action, checked on the backend and mirrored on the UI.

Lifecycle

Import, shape, secure, publish

The full path from a raw data source to a governed MCP endpoint an AI client can call, without leaving the browser.

01

Import

Connect a REST API, a SQL or NoSQL database, an OpenAPI/Swagger spec, or start from a blank server.

02

Shape

Turn endpoints or database operations into MCP tools, define resources and prompts, and set JSON Schema input/output per tool.

03

Secure

Store credentials as secrets, configure upstream auth, rate limits, and role-based permissions.

04

Publish & observe

Share a public MCP docs link, connect an AI client, and watch health, metrics, and audit logs.

Features

Everything you need to run MCP servers in production

One dashboard covering the full lifecycle of turning a data source into a governed MCP server.

MCP server management

Create servers from a REST spec, a database connection, a template, or scratch. Configure auth, rate limits, and maintenance mode per server, and pause traffic instantly without losing config.

69 REST API templates

Skip the blank-page problem. GitHub, Stripe, Slack, Notion, HubSpot, OpenAI, Shopify, Twilio, and dozens more, ready to wire up in one step.

Dynamic tools

Generate MCP tools automatically from imported endpoints or database operations, or hand-build them with full JSON Schema control and inline testing before shipping to an AI client.

Resources & prompts

Expose static or live-fetched resources, plus a shared prompt library with Handlebars-style arguments reused across every server.

Secrets & auth

Values stored separately from metadata, referenced by name instead of pasted inline. OAuth clients and MCP API keys managed alongside them.

Roles & permissions

Four built-in roles plus fully custom ones. Sensitive actions like revealing a secret get their own dedicated permission, never bundled into a broader one.

Public MCP docs pages

Generate a permanent, slug-based documentation link for one server's tools, resources, and prompts (never credentials), and revoke it any time.

Observability

Health, readiness, and liveness endpoints, structured JSON logs, Prometheus metrics, and optional OpenTelemetry tracing with a ready Grafana dashboard.

Audit logs & error tracking

A clean accountability trail for intentional changes, kept separate from runtime error monitoring wired to the provider your team already uses.

AI providers

Connect OpenAI or Anthropic to get AI assistance on tool names, descriptions, and schemas while importing a REST server. Credentials are tested before you rely on them.

Harness

Per-tool timeout limits, retry policy with backoff, and execution hooks. Configuration UI ships today, runtime enforcement is landing next.

Guard rails

Input constraints, output filtering, and tool restrictions to control exactly what an AI is allowed to send and receive through a tool call.