Enum Harness 

pub enum Harness {
    ClaudeCode,
    Codex,
    Cursor,
    Cline,
    Continue,
    Aider,
    Goose,
    ClaudeDesktop,
    Generic(String),
}

MCP harness detected from the initialize.clientInfo.name field.

The variants cover the harnesses called out in docs/v0.7/compatibility-matrix.html (Track D2). Unknown harnesses fall through to Generic(String) carrying the original clientInfo.name so downstream logging / metrics can attribute behaviour to the unrecognised client without losing the name.

serde uses snake_case so the wire shape matches the rest of the v3 capabilities document. Generic carries an inner string and serialises as {"generic": "<name>"} per serde’s default externally-tagged enum representation.

Variants

ClaudeCode

Anthropic Claude Code — supports deferred-tool registration via ToolSearch. The only first-class harness today where B1’s memory_load_family actually surfaces new tools mid-session.

Codex

OpenAI Codex CLI — eager-load only.

Cursor

Anysphere Cursor — MCP via stdio, eager-load only.

Cline

VS Code Cline extension — MCP via stdio, eager-load only.

Continue

Continue.dev (VS Code / JetBrains) — MCP via stdio, eager-load only.

Aider

Aider CLI pair-programmer — MCP via stdio, eager-load only.

Goose

Block / Square Goose — MCP via stdio, eager-load only.

ClaudeDesktop

Anthropic Claude Desktop — eager-load only.

Generic(String)

Unknown harness; carries the original clientInfo.name so the operator can grep logs for “this is the harness I forgot to register” without losing the identifying string.

Implementations

impl Harness

pub fn detect(client_name: &str) -> Self

Detect a harness from the clientInfo.name field of the MCP initialize request.

Matching is case-insensitive and fuzzy substring — the raw name is normalised by lower-casing and stripping the punctuation harnesses use as separators (-, _, , .) before comparison. So "claude-code", "Claude Code", "claude_code", and "CLAUDE.CODE" all detect as Harness::ClaudeCode.

Unknown names round-trip into Generic(<original>) preserving the input verbatim so logging / metrics keep a useful label.

pub fn supports_deferred_registration(&self) -> bool

Whether this harness exposes tools registered after the initial tools/list to the LLM mid-session.

true only for harnesses with documented deferred-tool registration support. Today that’s just Claude Code via its ToolSearch mechanism. Every other known harness eager-loads the manifest at session start and won’t surface a tool added later — so B1’s memory_load_family falls back to a “restart with --profile <family>” hint on those harnesses.

Default for unknown harnesses (Generic): false (conservative). It’s better to under-promise than to claim a tool will appear and have the conversation strand on a stale cached manifest.

Trait Implementations

impl Clone for Harness

fn clone(&self) -> Harness

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Harness

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for Harness

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Eq for Harness

impl PartialEq for Harness

fn eq(&self, other: &Harness) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for Harness

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for Harness