aprender_mcp/tools/
bench.rs

1//! `apr.bench` — M2 tool. Benchmark inference throughput (tok/s, latency percentiles).
2//!
3//! Wraps `apr bench <model> --json [--iterations N] [--max-tokens N] [--prompt X]`.
4
5#![allow(clippy::disallowed_methods)] // serde_json::json! macro expands to .unwrap() internally
6
7use crate::tools::subprocess::run_apr;
8use crate::types::{InputSchema, ToolCallResult, ToolDefinition};
9
10/// Tool name registered with MCP clients.
11pub const NAME: &str = "apr.bench";
12
13/// Return the MCP tool definition for `apr.bench`.
14///
15/// FALSIFY-MCP-008: the `inputSchema` is parsed from the build-time codegen
16/// constant `crate::schemas::APR_BENCH_SCHEMA`, which `build.rs` emits from
17/// `contracts/apr-mcp-tool-schemas-v1.yaml`. The contract is the single
18/// source of truth — the live `tools/list` response and the YAML must agree
19/// byte-for-byte after JSON canonicalization (asserted by
20/// `tests/falsify_mcp_008.rs`).
21#[must_use]
22pub fn bench_tool_definition() -> ToolDefinition {
23    let input_schema: InputSchema = serde_json::from_str(crate::schemas::APR_BENCH_SCHEMA).expect(
24        "FALSIFY-MCP-008: apr.bench codegen constant must parse as InputSchema; \
25             regenerate by editing contracts/apr-mcp-tool-schemas-v1.yaml and rebuilding",
26    );
27    ToolDefinition {
28        name: NAME.to_string(),
29        description: crate::schemas::APR_BENCH_DESCRIPTION.to_string(),
30        input_schema,
31    }
32}
33
34/// Execute `apr.bench` by spawning `apr bench <model> --json [...flags]`.
35#[must_use]
36pub fn call(args: &serde_json::Value) -> ToolCallResult {
37    let Some(model_path) = args.get("model_path").and_then(|v| v.as_str()) else {
38        return ToolCallResult::error("Missing required argument: model_path");
39    };
40
41    let mut owned: Vec<String> = vec![
42        "bench".to_string(),
43        model_path.to_string(),
44        "--json".to_string(),
45    ];
46
47    if let Some(n) = args.get("iterations").and_then(serde_json::Value::as_u64) {
48        owned.push("--iterations".to_string());
49        owned.push(n.to_string());
50    }
51    if let Some(n) = args.get("max_tokens").and_then(serde_json::Value::as_u64) {
52        owned.push("--max-tokens".to_string());
53        owned.push(n.to_string());
54    }
55    let prompt = args.get("prompt").and_then(|v| v.as_str()).unwrap_or("");
56    if !prompt.is_empty() {
57        owned.push("--prompt".to_string());
58        owned.push(prompt.to_string());
59    }
60
61    let argv: Vec<&str> = owned.iter().map(String::as_str).collect();
62    run_apr(&argv)
63}
64
65/// HELIX-IDEA-002 — unified-signature shim for the inventory dispatcher.
66pub fn dispatch(
67    args: &serde_json::Value,
68    _cancel: &std::sync::mpsc::Receiver<()>,
69    _sink: Option<&crate::server::NotificationSink>,
70    _token: Option<serde_json::Value>,
71) -> ToolCallResult {
72    call(args)
73}
74
75crate::register_mcp_tool!(
76    name: NAME,
77    definition: bench_tool_definition,
78    dispatch: dispatch,
79);
80
81#[cfg(test)]
82#[allow(clippy::disallowed_methods)]
83mod tests {
84    use super::*;
85
86    #[test]
87    fn definition_has_correct_name_and_required_field() {
88        let def = bench_tool_definition();
89        assert_eq!(def.name, "apr.bench");
90        assert_eq!(def.input_schema.schema_type, "object");
91        assert_eq!(def.input_schema.required, vec!["model_path".to_string()]);
92        for field in ["model_path", "iterations", "max_tokens", "prompt"] {
93            assert!(
94                def.input_schema.properties.contains_key(field),
95                "property {field} present"
96            );
97        }
98    }
99
100    #[test]
101    fn missing_model_path_returns_error() {
102        let result = call(&serde_json::json!({}));
103        assert_eq!(result.is_error, Some(true));
104        assert!(result.content[0].text.contains("model_path"));
105    }
106}
aprender_mcp/tools/bench.rs

aprender_mcp/tools/
bench.rs
