MCP + Agent Skill

McpBackend MCP Server and Agent Skill

Connect your AI coding tool to McpBackend's hosted MCP server, then install the open Agent Skill that teaches it a secure, repeatable backend workflow.

Updated July 17, 2026

Tools to act, a skill to act well

The hosted MCP server gives your coding agent 16 authenticated tools for projects, API contracts, schemas, tables, row-level security, API keys, end-user auth, webhooks, usage, and team access. The open Agent Skill adds the operating playbook: inspect live state first, make the smallest change, choose permissions explicitly, keep secrets out of source control, and verify the result.

  • MCP server

    The live control surface. It performs authorized operations against your McpBackend account through OAuth.

  • Agent Skill

    The portable workflow. It tells a compatible agent how to plan, secure, and verify McpBackend changes.

Your application does not use MCP at runtime. It keeps using the generated REST API; MCP and the skill are for the coding agent that builds and manages the backend.

Install the open McpBackend Agent Skill

The skill follows the open Agent Skills format and is published at github.com/mcpbackend/skills. Review the source, then install it with the skills CLI:

npx skills add https://github.com/mcpbackend/skills --skill mcpbackend

After installation, invoke it as $mcpbackend or ask for a task such as “create a secure McpBackend project for a client portal.” Compatible agents can also activate it automatically when the request matches its description.

Connect the hosted MCP server

Add the Streamable HTTP endpoint to your MCP client. The first connection opens McpBackend sign-in and consent in your browser.

{
  "mcpServers": {
    "mcpbackend": {
      "url": "https://mcp.mcpbackend.com/mcp"
    }
  }
}

For Claude Code, the equivalent command is:

claude mcp add --transport http mcpbackend https://mcp.mcpbackend.com/mcp

What a skill-guided agent does

  1. 1

    Clarify the app

    Establish entities, relationships, important queries, user authentication, access rules, and whether EU residency is required.

  2. 2

    Inspect live state

    List projects and read the current schema before assuming what already exists.

  3. 3

    Create the smallest schema

    Add intentional columns, relationships, uniqueness rules, and indexes for real query paths.

  4. 4

    Set security explicitly

    Enable auth when needed and choose public, authenticated, or owner-only reads and writes for every table.

  5. 5

    Issue scoped credentials

    Create least-privilege API keys only after access rules are settled and store one-time secrets outside source control.

  6. 6

    Verify and hand off

    Read the schema again, check usage, and summarize the backend without exposing secrets.

Security boundaries the workflow preserves

  • OAuth instead of pasted tokens

    The MCP client completes browser-based authorization and keeps its own credential.

  • Live inspection before writes

    The agent reads projects and schema instead of relying on stale chat context.

  • Least-privilege keys

    Per-table permissions are preferred over permanent full-account credentials.

  • Explicit row access

    Public, authenticated, and owner modes are chosen from the application's actual user flows.

  • One-time secrets stay secret

    API key and webhook secrets go to environment configuration, not commits or summaries.

  • Human intent for team changes

    Invitations and access revocation are performed only when explicitly requested.

Frequently asked questions

What is the difference between the MCP server and the Agent Skill?+

The MCP server provides authenticated tools that can inspect and change your McpBackend account. The Agent Skill is a portable instruction set that teaches compatible agents when and how to use those tools, including schema design, least-privilege access, and verification steps.

Do I need both?+

The MCP server works on its own. Installing the Agent Skill adds McpBackend-specific workflow guidance, so the agent is less likely to guess at schemas, permissions, or credential handling.

How does the MCP server authenticate?+

It uses OAuth 2.1. Your MCP client opens McpBackend sign-in and consent in the browser, then stores its own token. You do not paste a dashboard session or email code into configuration.

Where is the skill source code?+

The skill is published in the public mcpbackend/skills repository on GitHub, so you can inspect its instructions and tool reference before installing it.

Build the backend, skip the setup

Give your AI editor a real backend

Start with a free project. Database, REST API, auth, permissions, webhooks, and MCP are included.

Related guides