mcp-graphql-forge
About
A lightweight, configuration-driven MCP server that exposes curated GraphQL queries as modular tools, enabling intentional API interactions from your agents.
Details
- Author
- UnitVectorY-Labs
- GitHub stars
- 5
- Downloads
- 201
- Categories
- Other
Jump to
- Configuration-driven YAML tool definitions for modularity.
- Supports multiple output formats: raw, json, and TOON.
- Optional bearer token fetching via command.
- Annotations for tool metadata (read-only, idempotent, etc.).
- Streamable HTTP mode with --http flag.
- Pre-compiled binaries for all major platforms.
Install by downloading a pre-compiled executable from GitHub Releases for macOS, Linux, or Windows, or via go install github.com/UnitVectorY-Labs/mcp-graphql-forge@latest. Configure with the --forgeConfig command line parameter (or FORGE_CONFIG environment variable) pointing to a folder containing forge.yaml (server URL, optional token command) and tool definition YAML files. Run in stdio mode by default; add --http 8080 for streamable HTTP mode.
mcp-graphql-forge
A lightweight, configuration-driven MCP server that exposes curated GraphQL queries as modular tools, enabling intentional API interactions from your agents.
Purpose
mcp-graphql-forge lets you turn any GraphQL endpoint into an MCP server whose tools are defined in YAML files that specify the GraphQL queries and their parameters. This allows you to create a modular, secure, and minimal server that can be easily extended without modifying the application code.
Releases
All official versions of mcp-graphql-forge are published on GitHub Releases. Since this MCP server is written in Go, each release provides pre-compiled executables for macOS, Linux, and Windows—ready to download and run.
Alternatively, if you have Go installed, you can install mcp-graphql-forge directly from source using the following command:
go install github.com/UnitVectorY-Labs/mcp-graphql-forge@latest
Configuration
The server is configured using command line parameters, environment variables, and YAML files.
Command Line Parameters
- --forgeConfig: Specifies the path to the folder containing the YAML configuration files (forge.yaml and tool definitions). If set, this takes precedence over the FORGE_CONFIG environment variable. If neither is set, the application will return an error and exit.
- --forgeDebug: If provided, enables detailed debug logging to stderr, including the obtained token and the full HTTP request/response for GraphQL calls. If set, this takes precedence over the FORGE_DEBUG environment variable.
Environment Variables
- FORGE_CONFIG: Specifies the path to the folder containing the YAML configuration files (forge.yaml and tool definitions). Used if --forgeConfig is not set.
- FORGE_DEBUG: If set to true (case-insensitive), enables detailed debug logging to stderr, including the obtained token and the full HTTP request/response for GraphQL calls. Used if --forgeDebug is not set.
forge.yaml
The configuration folder uses a special configuration file forge.yaml that specifies the common configuration attributes.
The following attributes can be specified in the file:
- name: The name of the MCP server
- url: The URL of the GraphQL endpoint
- token_command: The command to use to request the Bearer token for the Authorization header (optional)
- env: A map of environment variables to pass to the token command (optional)
- env_passthrough: If set to true, passes all environment variables used when invoking mcp-graphql-forge to the token command; if used in conjunction with env, the variables from env will take precedence (optional, defaults to false)
An example configuration would look like:
name: "ExampleServer"
url: "https://api.github.com/graphql"
token_command: "gh auth token"
Tool Configuration
All other YAML files located in the folder are treated as configuration files. Each YAML file defines a tool for the MCP server.
The following attributes can be specified in the file:
- name: The name of the MCP tool
- description: The description of the MCP tool
- query: The GraphQL query to execute
- inputs: The list of inputs defined by the MCP tool and passed into the GraphQL query as variables
- name: The name of the input
- type: The parameter type; can be 'string' or 'number'
- description: The description of the parameter for the MCP tool to use
- required: Boolean value specifying if the attribute is required
- annotations: MCP annotations that provide hints about the tool's behavior (optional)
- title: A human-readable title for the tool, useful for UI display (optional)
- readOnlyHint: If true, indicates the tool does not modify its environment (optional, default: false)
- destructiveHint: If true, the tool may perform destructive updates (only meaningful when readOnlyHint is false) (optional, default: true)
- idempotentHint: If true, calling the tool repeatedly with the same arguments has no additional effect (only meaningful when readOnlyHint is false) (optional, default: false)
- openWorldHint: If true, the tool may interact with an "open world" of external entities (optional, default: true)
- output: The output format for the GraphQL response (optional, defaults to "raw")
- raw: Passes through the server response as-is (default)
- json: Returns minimized JSON with unnecessary spacing removed
- toon: Converts JSON response to TOON format (Token-Oriented Object Notation) for efficient token usage with LLMs
An example configuration would look like:
name: "getUser"
description: "Fetch basic information about a user by login, including their name, URL, and location."
query: |
query ($login: String!) {
user(login: $login) {
id
name
url
location
}
}
inputs:
- name: "login"
type: "string"
description: "The user login that uniquely identifies their account."
required: true
annotations:
title: "Get User Information"
readOnlyHint: true
destructiveHint: false
idempotentHint: true
openWorldHint: true
output: "toon" # Optional: "raw" (default), "json", or "toon"
Output Formats
The output configuration parameter allows you to specify how the GraphQL response should be formatted. This is particularly useful when working with LLMs, as different formats can optimize for token efficiency.
Available Output Formats
raw (default)
- Passes through the GraphQL server response exactly as received
- No transformation or modification is applied
- Use when you need the original formatting from the server
json
- Returns minimized JSON with all unnecessary whitespace removed
- Reduces token usage compared to formatted JSON
- Ideal for reducing payload size while maintaining JSON compatibility
- Automatically falls back to raw output if the response is not valid JSON
toon
- Converts the JSON response to TOON format (Token-Oriented Object Notation)
- Can reduce token usage by 30-60% for uniform, tabular data
- Optimized for LLM consumption with compact, human-readable output
- Uses the toon-format/toon-go library
- Automatically falls back to raw output if the response is not valid JSON or conversion fails
Run in Streamable HTTP Mode
By default the server runs in stdio mode, but if you want to run in streamable HTTP mode, you can specify the --http command line flag with the server address and port (ex: --http 8080). This will run the server with the following endpoint that your MCP client can connect to:
http://localhost:8080/mcp
./mcp-graphql-forge --http 8080
If you do not specify token_command in the configuration, the "Authorization" header, if passed to the MCP server, will be passed through from the incoming MCP request to the backend GraphQL endpoint.
Limitations
- Each instance of mcp-graphql-forge can only be used with a single GraphQL server at a single URL.
- The GraphQL queries are all exposed as Tools and not as Resources, even if they are not mutations.
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.



