mirror of
https://github.com/hanzoai/mcp.git
synced 2026-10-15 12:30:12 +00:00
zeekay
de99c2e61f
The kai section names hanzo-actions (rust/actions) as the executor, and the native-tools section says which crate Dev links for what: hanzo- trigram for search, hanzo-actions for actions, hanzo-mcp for ExecTool. fs, code, exec and kai together are 6.5 KB, about 1.6K tokens; the surface test holds them under 16 KB.
75 lines
3.1 KiB
Rust
75 lines
3.1 KiB
Rust
//! The tool surface a client sees: the same order every start, and as small as asked.
|
|
|
|
use hanzo_mcp::ToolRegistry;
|
|
use serde_json::Value;
|
|
|
|
fn names(registry: &ToolRegistry) -> Vec<String> {
|
|
registry
|
|
.get_definitions()
|
|
.iter()
|
|
.map(|d| d["name"].as_str().unwrap().to_string())
|
|
.collect()
|
|
}
|
|
|
|
#[test]
|
|
fn the_list_has_one_order_however_many_times_it_is_built() {
|
|
let first = names(&ToolRegistry::with_defaults());
|
|
for _ in 0..20 {
|
|
assert_eq!(first, names(&ToolRegistry::with_defaults()));
|
|
}
|
|
// The built-ins keep their fixed order; the registered tools follow, by name.
|
|
let custom = &first[16..];
|
|
let mut sorted = custom.to_vec();
|
|
sorted.sort();
|
|
assert_eq!(custom, sorted.as_slice());
|
|
assert_eq!(&first[..4], ["exec", "fs", "plan", "think"]);
|
|
}
|
|
|
|
#[test]
|
|
fn restricting_lists_and_serves_only_what_was_named() {
|
|
let mut registry = ToolRegistry::with_defaults();
|
|
registry.restrict(["fs", "code", "exec"]).unwrap();
|
|
assert_eq!(names(®istry), ["exec", "fs", "code"]);
|
|
assert_eq!(registry.list(), ["code", "exec", "fs"]);
|
|
}
|
|
|
|
#[tokio::test]
|
|
async fn a_tool_outside_the_list_is_unknown_not_hidden() {
|
|
let mut registry = ToolRegistry::with_defaults();
|
|
registry.restrict(["fs"]).unwrap();
|
|
let denied = registry.execute("exec", serde_json::json!({"action": "help"})).await.unwrap();
|
|
assert!(!denied.success);
|
|
assert_eq!(denied.error.as_deref(), Some("Unknown tool: exec"));
|
|
assert!(registry.execute("fs", serde_json::json!({"action": "help"})).await.unwrap().success);
|
|
// A registered tool is held to the list too.
|
|
let denied = registry.execute("code_search", serde_json::json!({"query": "x"})).await.unwrap();
|
|
assert_eq!(denied.error.as_deref(), Some("Unknown tool: code_search"));
|
|
}
|
|
|
|
#[test]
|
|
fn an_unknown_or_empty_list_is_an_error() {
|
|
let mut registry = ToolRegistry::with_defaults();
|
|
let err = registry.restrict(["fs", "serach"]).unwrap_err().to_string();
|
|
assert!(err.starts_with("unknown tool \"serach\""), "{err}");
|
|
assert!(ToolRegistry::with_defaults().restrict(["", " "]).is_err());
|
|
assert!(ToolRegistry::with_defaults().restrict(Vec::<String>::new()).is_err());
|
|
// A failed restrict leaves the registry whole.
|
|
assert!(registry.list().len() > 20);
|
|
}
|
|
|
|
/// What a client is charged each turn: the schemas of the tools it is shown. The search,
|
|
/// exec and code tools together stay under 4K tokens (16 KB at four bytes a token).
|
|
#[test]
|
|
fn the_native_tools_fit_in_four_thousand_tokens() {
|
|
let mut registry = ToolRegistry::with_defaults();
|
|
registry.restrict(["fs", "code", "exec", "kai_decide"]).unwrap();
|
|
let bytes: usize = registry.get_definitions().iter().map(|d| serde_json::to_string::<Value>(d).unwrap().len()).sum();
|
|
assert!(bytes < 16_000, "{bytes} bytes");
|
|
|
|
// With Kai Actions beside them.
|
|
let mut registry = ToolRegistry::with_defaults();
|
|
registry.restrict(["fs", "code", "exec", "kai"]).unwrap();
|
|
let bytes: usize = registry.get_definitions().iter().map(|d| serde_json::to_string::<Value>(d).unwrap().len()).sum();
|
|
assert!(bytes < 16_000, "{bytes} bytes with kai");
|
|
}
|