Local-first MCP server studio

Your data,
ready for AI.

Connect approved database queries and existing APIs to ChatGPT, Claude, Codex, and other MCP clients—without writing protocol code or handing your data to another platform.

Start with the guide →See how it works
Runs on your infrastructureSafe query controlsOpen MCP endpoint
Local runtime readyPHASE 05 · v1.0
RUNTIME / MCP ENDPOINT

Three tools, one secure URL.

SQLfind_customerSQL Server · read onlyLive
APIcreate_invoiceOpenAPI · confirmation requiredLive
MCP/mcp/customer-dataBearer protectedCopy
PostgreSQLMySQLSQLiteSQL ServerOpenAPISwaggerREST APIs
Everything in one control plane

From connection
to callable tool.

MCP Forge handles the technical translation while you decide exactly which operations an AI is permitted to use.

⌁

Database connections

Connect PostgreSQL, MySQL, SQLite, and SQL Server using familiar connection strings and dedicated service accounts.

API

OpenAPI importer

Paste Swagger or OpenAPI JSON/YAML. Operation names, parameters, descriptions, and request bodies become MCP tools.

SQL

Approved queries

Publish parameterized queries instead of unrestricted SQL. Enforce read-only statements and bounded result sets.

◆

Production policies

Apply redaction, write confirmation, timeouts, row limits, authentication, and explicit tool descriptions.

▷

Test before sharing

Inspect tools, validate JSON inputs, simulate calls, and view activity before connecting an AI client.

↯

Live MCP endpoints

Create stable HTTP endpoints supporting initialize, ping, tools/list, and tools/call with optional bearer protection.

A guided four-step setup

No protocol expertise
required.

Business users can follow the visual usage guide, while technical owners retain control of credentials and permissions.

01

Choose a source

Select your database type or paste an API specification.

02

Define allowed work

Add approved queries or review imported API operations.

03

Test the tools

Confirm names, parameters, results, and safety policies.

04

Share one endpoint

Copy the protected MCP URL into your chosen AI client.

Supported sources

Built around the
systems you already use.

View every setup guide →
PostgreSQLMySQLSQLiteMicrosoft SQL ServerOpenAPI 3.xSwaggerREST APIsDocker
Frequently asked questions

Clear answers,
before you begin.

What is an MCP server?

An MCP server gives an AI client a controlled list of tools it can call. MCP Forge translates approved queries and API operations into those tools.

Does MCP Forge upload my database?

No. MCP Forge runs locally or on infrastructure you control. Database traffic flows between your MCP Forge runtime and your configured database.

Can an AI run arbitrary SQL?

Not through the approved-query workflow. You define the exact parameterized SQL statement and its named inputs. Read-only enforcement is enabled by default.

Which databases are supported?

PostgreSQL, MySQL, SQLite, and Microsoft SQL Server are supported in the current runtime.

Can I import Swagger or OpenAPI?

Yes. Import JSON or YAML in Runtime → API Specs. Each supported operation becomes a tool using its operationId, parameters, request body, and description.

How should I secure a public endpoint?

Enable MCPFORGE_TOKEN, use HTTPS through a reverse proxy, restrict network access, and use database accounts with the minimum necessary permissions.

Where can I get help?

Email support@builtbykris.com.

Build your first MCP server.

The step-by-step guide starts with the information you already have.

Open the usage guide →