Masterytrace

What masterytrace MCP server does

The masterytrace MCP server connects an MCP client to the installed MasteryTrace command-line interface. It provides one tool, run, whose input is an argument array and whose result is a parsed dictionary. Because the tool forwards arguments to the CLI, an agent can reach the supported subcommands and flags through the same interface used from a terminal.

MasteryTrace tracks learner performance by skill. Its scoring engine supports Bayesian Knowledge Tracing (BKT), which produces posterior mastery probabilities, and two-parameter logistic Item Response Theory (2PL IRT), which produces ability estimates and item-related parameters. The CLI can run either model or both.

How it works

The server shells out to the installed masterytrace executable. A call such as run(["score", "--model", "bkt", "--json"]) asks the CLI to score the stored event log with BKT and return JSON-compatible output. The global --json option is useful for agent calls because it selects machine-readable output; report also supports table, JSON, and Markdown formats.

The underlying workflow stores state locally in a .masterytrace directory. init creates sample input and configuration files, record validates and stores a JSON or CSV event log, score writes model results, and report reads those results for display. The event records identify a learner, skill, correctness value, and ISO 8601 timestamp.

Setup and configuration

Install the Python distribution with its MCP extra:

pip install "masterytrace-cli[mcp]"

This provides the masterytrace-mcp console script and the masterytrace CLI. Configure the client to launch that script over its standard input/output connection. For Claude Desktop, the server entry is:

{
  "mcpServers": {
    "masterytrace": {
      "command": "masterytrace-mcp"
    }
  }
}

The working directory matters because the CLI reads and writes .masterytrace state relative to where it runs. No database or service endpoint is required, and the README does not specify environment variables for this server.

Tools and capabilities

The masterytrace MCP server exposes only run, rather than separate MCP tools for each operation. Through its argument list, agents can:

• Scaffold sample event and configuration files with init.
• Load and validate JSON or CSV event logs with record.
• Score stored events using BKT, IRT, or both with score.
• Produce table, JSON, or Markdown reports with report.
• Request JSON output and use documented exit codes for success, usage errors, and invalid event data.

record replaces the existing stored event log instead of appending to it, so callers that add responses must submit a merged event file.

Limitations and notes

This is a generic CLI wrapper: the server does not add domain operations beyond the commands and flags supported by MasteryTrace. The event log must follow the documented schema, and files larger than 100 MB are rejected. JSON and CSV inputs use slightly different field conventions, including learnerId and skillId for JSON versus learner_id and skill_id for CSV.

The npm and Python MasteryTrace distributions produce equivalent data but use different JSON key casing. The MCP setup described here uses the Python package. Since state is local, the client process must have access to the working directory containing the relevant .masterytrace files.