@koda-sl/baker-bridge 0.109.0-dev.f28c07bd2 → 0.109.0-dev.fbf7b641a

package/README.md

@@ -48,6 +48,7 @@ await server.start();
 | `BAKER_ACTING_USER_ID` | Injected into the Agent SDK subprocess env per turn (when the turn's sender is known) so in-sandbox `baker` CLI calls can scope work to the acting user — e.g. `baker mcp add --scope user` |
 | `CLAUDE_CODE_ENABLE_TASKS` | Set to `0` on every Agent SDK subprocess. Turns off the CLI's built-in task backlog and its `task_reminder` — see [Session surface](#session-surface) |
 | `CLAUDE_CODE_MODEL_CAPABILITIES` | Set to `-per_turn_effort` on every Agent SDK subprocess so no request carries a per-message effort, which OpenRouter cannot route to Bedrock — see [Bedrock is the preferred provider](#bedrock-is-the-preferred-provider) |
+| `ENABLE_TOOL_SEARCH` | Set to `true` on every Agent SDK subprocess. MCP tool schemas are deferred to their names until the agent loads one with `ToolSearch`; the CLI would otherwise load every schema into every request, because it switches tool search off behind a non-first-party `ANTHROPIC_BASE_URL` — see [MCP servers (Composio)](#mcp-servers-composio) |
 | `NO_COLOR` | Set to `1` on every Agent SDK subprocess so `baker --help` output carries no ANSI escapes — see [Session surface](#session-surface) |
 | `BAKER_AGENT_MODEL` | Optional. Runs every turn in this runtime on the named model instead of the default, and is forwarded to subagents as `CLAUDE_CODE_SUBAGENT_MODEL`. Ignored unless it names a model with a known context window — see [Model selection](#model-selection) |
 | `CLAUDE_CODE_SUBAGENT_MODEL`, `CLAUDE_CODE_SUBAGENT_MODEL_FORCE` | Set on the Agent SDK subprocess only when `BAKER_AGENT_MODEL` is set. Together they override the `model:` each subagent pins in its own frontmatter |
@@ -284,6 +285,8 @@ When the agent *does* call `request_avatar_casting` for the avatar a pending cas
 
 Before each turn the bridge fetches the company's scoped Composio MCP endpoints from Convex — `POST {BAKER_CONVEX_SITE_URL}/api/chat/mcp-servers` with the bridge's `BAKER_API_KEY` bearer, sending `{ threadId, actingUserId? }`. The response `{ mcpServers }` map (up to four streamable-HTTP entries: `composio_user`, `composio_user_org`, `composio_company`, `composio_org` — plus any custom servers) is passed straight into `options.mcpServers`. No Composio secret is injected into the sandbox env; the capability URLs live only in the bridge process.
 
+Connected servers are deferred behind tool search (`ENABLE_TOOL_SEARCH=true`): each request carries their tool names, and a tool's schema enters the prompt only once the agent searches for it. A large toolkit (Make exposes ~180 tools) therefore costs its names, not ~115k tokens on every request. `baker_ui` is the exception — it sets `alwaysLoad`, so its forms are always one call away.
+
 The fetch is resilient by design: results are cached per `(thread, actor)` for 60s, and a failed or timed-out refetch (2.5s) serves the last set that `(thread, actor)` received, however old — the tool list is part of the prompt, so dropping it mid-chat would rewrite the prompt cache. Only an actor that never fetched successfully in that chat runs the turn with **no** MCP; nothing here fails a turn. The actor is part of the key because personal connections (`user`, `user_org`, per-user custom-server sign-ins) belong to the person sending the message, so a teammate's turn never reuses them. When the Agent SDK reports an MCP server failed to connect, the bridge fires `POST /api/chat/mcp-invalid`: a `composio_*` entry is reported by scope, so Convex re-mints that scope's session on the next turn, and a custom server is reported by name (`customServerName`), so Convex re-checks its address and records why it failed instead of the entry being silently dropped.
 
 ### Tool mentions

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@koda-sl/baker-bridge",
-  "version": "0.109.0-dev.f28c07bd2",
+  "version": "0.109.0-dev.fbf7b641a",
   "description": "HTTP server wrapping the Claude Agent SDK. Persists chat events to Convex; the dashboard subscribes via Convex realtime",
   "license": "MIT",
   "type": "module",