MOSTLY AI

You run an MCP server that lets LLM agents securely interact with the MOSTLY AI Platform using OAuth 2.1. It exposes a set of endpoints and a local runtime you can start on your machine, enabling remote clients to query platform information, manage connectors, generate synthetic data, and more through a standard MCP interface.

How to use

You connect an MCP client to the server using either a remote MCP URL or a local runtime that you start on your machine. Once connected, you can browse available models, generators, and synthetic datasets, read and write connector data, run read-only queries, and monitor long-running tasks like generator training or synthetic data generation. The server supports OAuth 2.1 for agent authentication, so ensure your client is configured to present valid access tokens when making requests.

How to install

Prerequisites you need before installation: a system with a supported runtime for MCP and network access to reach the MCP host or run it locally. The runtime is started via a short command, and you should prepare any required environment variables before starting.

Option 1: Run remotely via a hosted MCP URL (recommended for quick start) You can connect to the remote MCP endpoint at the following address to start using the server without local installation.

Option 2: Run locally on your machine You install the server runtime and start it with a single command. The final start command is shown here so you can copy-paste it into your shell.

Note: You should set up the necessary environment variables and any required configuration files before starting the server locally.

Additional notes

Configuration and runtime specifics appear in the sections below. The server provides an HTTP connection method for remote use and a local stdio method for quick development testing. When you run locally, the server listens on port 8000 unless you override it.

Security considerations include using OAuth 2.1 tokens for client requests and protecting secret keys. For debugging and testing, you can use the MCP Inspector tool to validate endpoints and flows.

Usage flows supported by the server include querying platform information, listing and inspecting models and computes, managing connectors, running read-only connector queries, and orchestrating generator or synthetic dataset tasks with progress polling.

Configuration examples and runtime commands

{
  "type": "http",
  "name": "mostlyai",
  "url": "https://mcp.mostly.ai/sse",
  "args": []
}

{
  "type": "stdio",
  "name": "mostlyai",
  "command": "uv",
  "args": ["run", "mostlyai-mcp-server", "--port", "8000"]
}

Available tools

get_platform_info

retrieve information about the platform

get_user_info

retrieve information about the current user

get_models

retrieve a list of available models of a specific type

get_computes

retrieve a list of available compute resources for tasks

list_connectors

list all available connectors

get_connector

get details of a specific connector

create_connector

create a new connector and optionally test the connection

get_connector_locations

list available locations (schemas, tables, buckets, etc.) for a connector

get_connector_schema

get the schema of a table or file at a connector location

read_connector_data

read data from a connector location

write_connector_data

write data to a connector location (currently disabled)

query_connector

execute a read-only SQL query against a connector's data source

list_generators

list all available generators

get_generator

get details of a specific generator

poll_generator_progress

poll progress of a generator training job

train_generator

train a new generator with provided data or configuration

clone_generator

clone an existing generator

get_synthetic_dataset

get details of a specific synthetic dataset

poll_synthetic_dataset_progress

poll progress of a synthetic dataset generation job

generate_synthetic_dataset

generate a synthetic dataset using a generator

list_synthetic_datasets

list all available synthetic datasets

probe_generator

probe a generator for a new synthetic dataset (quick sample)