perf(agent): cache the MCP tool list per client (F1)

Fix F1 of plans/2026-07-11-nomos-agent-code-review.md, the last item.
buildTools called listToolsFull (a tools/list MCP round-trip) at the start of
EVERY chat turn, including every auto-continuation resume — the tool list is
static for the lifetime of one MCP connection, changing only when the api
process re-registers tools (a restart, which this client already detects and
reacts to via reconnectLocked). Re-fetching it every single turn was
avoidable network+parsing work on the hot path.

mcpClient now caches the parsed tool list after its first fetch, guarded by
its own mutex (kept separate from the request-serializing mu so a cache
check never contends with an in-flight doRequest call). reconnectLocked
clears the cache — an api restart may have changed what's registered, so a
stale cache would be wrong, not just slow. fleetSnapshot's get_health_summary
call is deliberately left uncached — it's meant to be "as of now."

Since each session gets its own client (the per-session pool from the
concurrency work), this caches per-task-conversation rather than globally: a
task's FIRST turn still pays the round-trip, every turn after reuses the
cached list — which is exactly the case that mattered (long-running,
heavily-autonomous tasks with many auto-continuation resumes).

Verified live via the api's request log: a brand-new session's first turn
made 3 MCP calls (initialize, tools/list, get_health_summary); a second turn
on the SAME session made exactly 1 (only get_health_summary) — tools/list
correctly skipped.

This completes the implementation order in
plans/2026-07-11-nomos-agent-code-review.md — every A/B/D/E/F finding from
the review (excluding C1, explicitly deferred per operator instruction) is
now fixed, tested, and verified live.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-07-11 20:30:07 +02:00
parent 6d4f6de676
commit 11c18e8956
2 changed files with 38 additions and 0 deletions

View File

@@ -481,6 +481,18 @@ type mcpClient struct {
http *http.Client
nextID int
mu sync.Mutex // this client is one stateful MCP session; serialize ITS OWN calls
// toolsCache holds the last tools/list result. The tool list is static
// for the lifetime of one MCP connection — it only changes when the api
// process (re)registers tools, i.e. on a restart, which this client
// already detects and reacts to via reconnectLocked. Without this,
// buildTools (called at the start of EVERY chat turn, including every
// auto-continuation resume) paid a full tools/list round-trip every
// single time for a list that's almost always identical to the last one.
// Guarded separately from mu (not reused) so a cache check never
// contends with an in-flight doRequest call for a different method.
toolsMu sync.Mutex
toolsCache []toolDef
}
func newMCPClient(baseURL string) (*mcpClient, error) {
@@ -540,6 +552,12 @@ func (c *mcpClient) doRequest(method string, params map[string]any) (*mcpJSONRPC
// reconnectLocked re-initializes the MCP session. The caller must hold c.mu.
func (c *mcpClient) reconnectLocked() error {
c.sessionID = ""
// A reconnect means the api process was restarted (or forgot us) — its
// tool registration may have changed, so the cached list is no longer
// trustworthy.
c.toolsMu.Lock()
c.toolsCache = nil
c.toolsMu.Unlock()
resp, err := c.send("initialize", map[string]any{
"protocolVersion": "2024-11-05",
"capabilities": map[string]any{},