Create a new, independent AI red-team engagement and return its id.

The returned engagement id must be used for all subsequent target/attack/ scoring/report tool calls that operate within this engagement.

List all engagements, optionally filtered by status ('active' or 'closed').

Get the current summary state of a single engagement by id.

Close an engagement, cancelling any attacks still running within it.

Configure (or replace) the REST chat target for an engagement.

    `target.request_template` is the JSON body sent to the endpoint, with the
    literal token `target.prompt_placeholder` (default "{{PROMPT}}") substituted
    with the outgoing prompt text anywhere it appears. `target.response_text_path`
    is a dotted/indexed path used to extract the reply text from the JSON response
    (e.g. "choices.0.message.content").
    

Get the currently configured target for an engagement (secrets redacted).

Send a lightweight connectivity check to the engagement's configured target.

List available PyRIT attack strategies and their constructor parameters.

Discovery only -- this does not recommend or select a strategy. The calling agent must choose which attack to run based on its own plan.

Run a named PyRIT attack strategy against the engagement's configured target.

    Starts a new conversation. Use `continue_attack` to keep talking within the
    same conversation. Optionally scores the final response with `scorer_names`.
    

Continue a previous attack conversation with a new objective/turn.

Cancel a currently running attack within an engagement.

Score arbitrary text (e.g. a target response) with a named PyRIT scorer.

    Returns success/confidence/risk/explanation/raw_score/metadata per PRD contract.
    Scorers that require an LLM judge (see list_capabilities) reuse the engagement's
    configured target as the judge model.
    

List all scores recorded so far within an engagement.

Generate a report (format: json, markdown, or html) for an engagement.

    The report includes an executive summary, findings, scores, metadata, and
    a timeline of engagement events. The file is written under the server's
    report output directory and tracked on the engagement.
    

List all reports generated so far for an engagement.

Copy a previously generated report to an arbitrary destination path.

Get PyRIT MCP server identity, version, and current engagement counts.

Get the installed PyRIT framework version.

List discoverable server capabilities: attack strategies, scorers, datasets, report formats.

    Discovery only. The agent is responsible for deciding which capability to use.
    

Record a finding on an engagement (e.g. a confirmed jailbreak or policy violation).

Findings are included in generated reports. The agent decides what constitutes a finding; the server only stores it.

List all findings recorded so far on an engagement.

Basic liveness check for the MCP server.