@gtrabanco/pi-nan-provider 0.8.0 → 0.10.0

package/scripts/models.generated.ts

@@ -1,7 +1,7 @@
 // This file is auto-generated by scripts/generate-models.ts
 // Do not edit manually — run `bun run generate-models` to update.
 //
-// Source: https://models.dev/api.json (provider "nan"), fetched 2026-09-27T21:25:24.946Z
+// Source: https://models.dev/api.json (provider "nan"), fetched 2026-09-29T22:08:21.449Z
 // Provenance: every contextWindow/maxTokens/input/cost value traces to
 // models.dev or to the per-entry notes below. Nothing is invented; entries
 // models.dev documents incompletely are omitted and flagged instead.
@@ -71,7 +71,8 @@ export const NAN_GENERATED_MODELS: readonly GeneratedModelEntry[] = [
 			"cost": {
 				"input": 0,
 				"output": 0
-			}
+			},
+			"canonical_model_id": "deepseek/deepseek-v4.1-flash"
 		}
 	},
 	{
@@ -117,7 +118,14 @@ export const NAN_GENERATED_MODELS: readonly GeneratedModelEntry[] = [
 			"reasoning": true,
 			"reasoning_options": [
 				{
-					"type": "toggle"
+					"type": "effort",
+					"values": [
+						"none",
+						"low",
+						"medium",
+						"high",
+						"max"
+					]
 				}
 			],
 			"tool_call": true,
@@ -142,7 +150,8 @@ export const NAN_GENERATED_MODELS: readonly GeneratedModelEntry[] = [
 			"cost": {
 				"input": 0,
 				"output": 0
-			}
+			},
+			"canonical_model_id": "google/gemma-4-26b-a4b-it"
 		}
 	},
 	{
@@ -184,7 +193,17 @@ export const NAN_GENERATED_MODELS: readonly GeneratedModelEntry[] = [
 			"family": "glm-flash",
 			"attachment": true,
 			"reasoning": true,
-			"reasoning_options": [],
+			"reasoning_options": [
+				{
+					"type": "effort",
+					"values": [
+						"low",
+						"medium",
+						"high",
+						"max"
+					]
+				}
+			],
 			"tool_call": true,
 			"structured_output": true,
 			"temperature": true,
@@ -207,12 +226,13 @@ export const NAN_GENERATED_MODELS: readonly GeneratedModelEntry[] = [
 			"cost": {
 				"input": 0,
 				"output": 0
-			}
+			},
+			"canonical_model_id": "zhipuai/glm-5.3-flash"
 		}
 	},
 	{
-		"id": "mimo-v2.5",
-		"name": "MiMo-V2.5",
+		"id": "mimo-v2.6-flash",
+		"name": "MiMo-V2.6-Flash",
 		"reasoning": true,
 		"input": [
 			"text",
@@ -235,21 +255,21 @@ export const NAN_GENERATED_MODELS: readonly GeneratedModelEntry[] = [
 			"maxTokensField": "max_tokens"
 		},
 		"notes": [
-			"compat matches the maintainer's working ~/.pi/agent/models.json LiteLLM config for api.nan.builders (2026-09-04): supportsDeveloperRole false, supportsReasoningEffort true, maxTokensField max_tokens. NaN's docs example sets only supportsDeveloperRole: true and is not battle-tested. supportsFinishReason true (2026-09-13, issue #2): the LiteLLM gateway intermittently closes SSE streams before emitting finish_reason; with true pi-ai raises 'Stream ended without finish_reason', which matches pi-ai's retryable-provider pattern ('ended without') and is retried automatically, whereas false silently synthesized stop/toolUse and stalled the turn mid-answer. supportsUsageInStreaming true (2026-09-16, issue #7): NaN's published schema is silent about stream_options, but the live gateway honors it — two identical streaming calls per model, differing only in stream_options: { include_usage: true }, returned 0 usage chunks without it and exactly 1 with it (prompt/completion/reasoning/cached token counts) on deepseek-v4-flash, glm5.3-flash, qwen3.6, mimo-v2.5 and gemma4, and a real pi session then recorded token counts where it recorded zeros. pi-ai only sends stream_options when this is not false, and the sanitizer forwards it when the model declares true, so chat models opt in by default and usage is reported (issue #4). A model that does not report streaming usage can still opt out per model with a models.json compat override (supportsUsageInStreaming: false); the sanitizer then strips stream_options and the payload stays strict."
+			"compat matches the maintainer's working ~/.pi/agent/models.json LiteLLM config for api.nan.builders (2026-09-04): supportsDeveloperRole false, supportsReasoningEffort true, maxTokensField max_tokens. NaN's docs example sets only supportsDeveloperRole: true and is not battle-tested. supportsFinishReason true (2026-09-13, issue #2): the LiteLLM gateway intermittently closes SSE streams before emitting finish_reason; with true pi-ai raises 'Stream ended without finish_reason', which matches pi-ai's retryable-provider pattern ('ended without') and is retried automatically, whereas false silently synthesized stop/toolUse and stalled the turn mid-answer. supportsUsageInStreaming true (2026-09-16, issue #7): NaN's published schema is silent about stream_options, but the live gateway honors it — two identical streaming calls per model, differing only in stream_options: { include_usage: true }, returned 0 usage chunks without it and exactly 1 with it (prompt/completion/reasoning/cached token counts) on deepseek-v4-flash, glm5.3-flash, qwen3.6, mimo-v2.5 and gemma4, and a real pi session then recorded token counts where it recorded zeros. pi-ai only sends stream_options when this is not false, and the sanitizer forwards it when the model declares true, so chat models opt in by default and usage is reported (issue #4). A model that does not report streaming usage can still opt out per model with a models.json compat override (supportsUsageInStreaming: false); the sanitizer then strips stream_options and the payload stays strict.",
+			"manual-only: \"mimo-v2.6-flash\" skipped: models.dev provider \"nan\" now lists it, so the manual-only entry is not emitted and the models.dev values are used (omnimodal model (text, image, audio input) served by NaN; limits and modalities from https://nan.builders/docs/models#mimo-v2-6-flash (checked 2026-09-25, quotas re-checked 2026-09-29): 1,048,576 context / 131,072 max output / 1.0B monthly quota per member. Reasoning: accepted, depth not adjustable (docs table, checked 2026-09-29). Tool calling: yes. Streaming: yes. Input modalities (pi-representable): text, image (audio not representable in pi's Model type). Fallback entry: emitted only while models.dev provider nan does not list the model — models.dev started listing it on 2026-09-29, so the generator then keeps the models.dev values and attaches this note to that entry instead of emitting a second one.)"
 		],
 		"extras": {
-			"id": "mimo-v2.5",
-			"name": "MiMo-V2.5",
-			"description": "Open MiMo model for multimodal coding agents and long-context automation",
+			"id": "mimo-v2.6-flash",
+			"name": "MiMo-V2.6-Flash",
+			"description": "MiMo Flash model for multimodal coding agents and long-context automation",
 			"family": "mimo",
 			"attachment": true,
 			"reasoning": true,
 			"reasoning_options": [],
 			"tool_call": true,
 			"temperature": true,
-			"knowledge": "2024-12",
-			"release_date": "2026-04-22",
-			"last_updated": "2026-04-22",
+			"release_date": "2026-09-22",
+			"last_updated": "2026-09-22",
 			"modalities": {
 				"input": [
 					"text",
@@ -268,38 +288,10 @@ export const NAN_GENERATED_MODELS: readonly GeneratedModelEntry[] = [
 			"cost": {
 				"input": 0,
 				"output": 0
-			}
+			},
+			"canonical_model_id": "xiaomi/mimo-v2.6-flash"
 		}
 	},
-	{
-		"id": "mimo-v2.6-flash",
-		"name": "mimo-v2.6-flash",
-		"reasoning": true,
-		"input": [
-			"text",
-			"image"
-		],
-		"cost": {
-			"input": 0,
-			"output": 0,
-			"cacheRead": 0,
-			"cacheWrite": 0
-		},
-		"contextWindow": 1048576,
-		"maxTokens": 131072,
-		"compat": {
-			"supportsDeveloperRole": false,
-			"supportsReasoningEffort": true,
-			"supportsUsageInStreaming": true,
-			"supportsFinishReason": true,
-			"maxTokensField": "max_tokens"
-		},
-		"notes": [
-			"compat matches the maintainer's working ~/.pi/agent/models.json LiteLLM config for api.nan.builders (2026-09-04): supportsDeveloperRole false, supportsReasoningEffort true, maxTokensField max_tokens. NaN's docs example sets only supportsDeveloperRole: true and is not battle-tested. supportsFinishReason true (2026-09-13, issue #2): the LiteLLM gateway intermittently closes SSE streams before emitting finish_reason; with true pi-ai raises 'Stream ended without finish_reason', which matches pi-ai's retryable-provider pattern ('ended without') and is retried automatically, whereas false silently synthesized stop/toolUse and stalled the turn mid-answer. supportsUsageInStreaming true (2026-09-16, issue #7): NaN's published schema is silent about stream_options, but the live gateway honors it — two identical streaming calls per model, differing only in stream_options: { include_usage: true }, returned 0 usage chunks without it and exactly 1 with it (prompt/completion/reasoning/cached token counts) on deepseek-v4-flash, glm5.3-flash, qwen3.6, mimo-v2.5 and gemma4, and a real pi session then recorded token counts where it recorded zeros. pi-ai only sends stream_options when this is not false, and the sanitizer forwards it when the model declares true, so chat models opt in by default and usage is reported (issue #4). A model that does not report streaming usage can still opt out per model with a models.json compat override (supportsUsageInStreaming: false); the sanitizer then strips stream_options and the payload stays strict.",
-			"manual-only: \"mimo-v2.6-flash\" not yet on models.dev provider nan (omnimodal model (text, image, audio input) served by NaN, not yet listed on models.dev provider nan (checked 2026-09-25); included so the live /models refresh does not hand it UNKNOWN_MODEL_LIMITS. Same limits as mimo-v2.5: 1,048,576 / 131,072 / 1.0B monthly quota per member. Reasoning: accepted, depth not adjustable. Tool calling: yes. Streaming: yes. Input modalities (pi-representable): text, image (audio not representable in pi's Model type).)."
-		],
-		"extras": {}
-	},
 	{
 		"id": "qwen3.6",
 		"name": "Qwen3.6 35B-A3B",
@@ -343,7 +335,14 @@ export const NAN_GENERATED_MODELS: readonly GeneratedModelEntry[] = [
 			"reasoning": true,
 			"reasoning_options": [
 				{
-					"type": "toggle"
+					"type": "effort",
+					"values": [
+						"none",
+						"low",
+						"medium",
+						"high",
+						"max"
+					]
 				}
 			],
 			"tool_call": true,
@@ -368,7 +367,8 @@ export const NAN_GENERATED_MODELS: readonly GeneratedModelEntry[] = [
 			"cost": {
 				"input": 0,
 				"output": 0
-			}
+			},
+			"canonical_model_id": "alibaba/qwen3.6-35b-a3b"
 		}
 	},
 	{
@@ -428,7 +428,8 @@ export const NAN_GENERATED_MODELS: readonly GeneratedModelEntry[] = [
 			"cost": {
 				"input": 0,
 				"output": 0
-			}
+			},
+			"canonical_model_id": "alibaba/qwen3.8-flash"
 		}
 	}
 ];
@@ -436,15 +437,17 @@ export const NAN_GENERATED_MODELS: readonly GeneratedModelEntry[] = [
 export const GENERATED_CATALOG_META = {
 	source: "https://models.dev/api.json",
 	modelsDevProvider: "nan",
-	fetchedAt: "2026-09-27T21:25:24.946Z",
-	modelCount: 7,
-	models: ["deepseek-v4-flash","gemma4","glm5.3-flash","mimo-v2.5","mimo-v2.6-flash","qwen3.6","qwen3.8-flash"],
+	fetchedAt: "2026-09-29T22:08:21.449Z",
+	modelCount: 6,
+	models: ["deepseek-v4-flash","gemma4","glm5.3-flash","mimo-v2.6-flash","qwen3.6","qwen3.8-flash"],
 	notes: [
+	"non-chat: \"qwen-image-2.1\" excluded from the chat catalog (image-generation model (text→image; output modality image) documented in https://nan.builders/docs/models and https://nan.builders/openapi.json (checked 2026-09-29) — not a chat model, out of the static chat catalog (models.dev reports limit.output 0))",
 	"live-only: \"glm5.3\" kept out of the static catalog (premium-tier model (models.dev now documents it with 1M context / 131,072 max output; NaN docs https://nan.builders/docs/models + https://nan.builders/openapi.json, checked 2026-09-13) kept live-only so a non-premium key never sees a model it cannot call when the live /models fetch is unavailable; premium keys still get it via the /models refresh with conservative placeholder limits)",
 	"provider-removed: \"glm5.2\" excluded from the catalog (removed by NaN (2026-09-05); absent from the official chat model list in https://nan.builders/openapi.json and https://nan.builders/docs/models (checked 2026-09-07) while models.dev provider nan still listed it — excluded so regeneration does not resurrect it)",
+	"provider-removed: \"mimo-v2.5\" excluded from the catalog (removed by NaN (checked 2026-09-29): zero mentions in https://nan.builders/docs/models and in the model param description of https://nan.builders/openapi.json, and models.dev provider nan dropped it as well — superseded by mimo-v2.6-flash; excluded so regeneration does not resurrect it)",
 	"compat matches the maintainer's working ~/.pi/agent/models.json LiteLLM config for api.nan.builders (2026-09-04): supportsDeveloperRole false, supportsReasoningEffort true, maxTokensField max_tokens. NaN's docs example sets only supportsDeveloperRole: true and is not battle-tested. supportsFinishReason true (2026-09-13, issue #2): the LiteLLM gateway intermittently closes SSE streams before emitting finish_reason; with true pi-ai raises 'Stream ended without finish_reason', which matches pi-ai's retryable-provider pattern ('ended without') and is retried automatically, whereas false silently synthesized stop/toolUse and stalled the turn mid-answer. supportsUsageInStreaming true (2026-09-16, issue #7): NaN's published schema is silent about stream_options, but the live gateway honors it — two identical streaming calls per model, differing only in stream_options: { include_usage: true }, returned 0 usage chunks without it and exactly 1 with it (prompt/completion/reasoning/cached token counts) on deepseek-v4-flash, glm5.3-flash, qwen3.6, mimo-v2.5 and gemma4, and a real pi session then recorded token counts where it recorded zeros. pi-ai only sends stream_options when this is not false, and the sanitizer forwards it when the model declares true, so chat models opt in by default and usage is reported (issue #4). A model that does not report streaming usage can still opt out per model with a models.json compat override (supportsUsageInStreaming: false); the sanitizer then strips stream_options and the payload stays strict.",
 	"input includes image: NaN serves the Vision-Exp variant ('takes images as input', https://nan.builders/docs/models; the image_url content-parts in https://nan.builders/openapi.json list deepseek-v4-flash among the vision models). models.dev provider nan also lists text+image now (DeepSeek V4.1 Flash entry, checked 2026-09-13; its 2026-09-07 snapshot listed text only), so this override is kept as a pin for the vision capability rather than as a divergence. reasoning_effort_values=[]: the model decides per-request how much to reason (https://nan.builders/docs/models, checked 2026-09-25).",
-	"manual-only: \"mimo-v2.6-flash\" not yet on models.dev provider nan (omnimodal model (text, image, audio input) served by NaN, not yet listed on models.dev provider nan (checked 2026-09-25); included so the live /models refresh does not hand it UNKNOWN_MODEL_LIMITS. Same limits as mimo-v2.5: 1,048,576 / 131,072 / 1.0B monthly quota per member. Reasoning: accepted, depth not adjustable. Tool calling: yes. Streaming: yes. Input modalities (pi-representable): text, image (audio not representable in pi's Model type).).",
+	"manual-only: \"mimo-v2.6-flash\" skipped: models.dev provider \"nan\" now lists it, so the manual-only entry is not emitted and the models.dev values are used (omnimodal model (text, image, audio input) served by NaN; limits and modalities from https://nan.builders/docs/models#mimo-v2-6-flash (checked 2026-09-25, quotas re-checked 2026-09-29): 1,048,576 context / 131,072 max output / 1.0B monthly quota per member. Reasoning: accepted, depth not adjustable (docs table, checked 2026-09-29). Tool calling: yes. Streaming: yes. Input modalities (pi-representable): text, image (audio not representable in pi's Model type). Fallback entry: emitted only while models.dev provider nan does not list the model — models.dev started listing it on 2026-09-29, so the generator then keeps the models.dev values and attaches this note to that entry instead of emitting a second one.)",
 	"contextWindow 262,144: the earlier 1,000,000 override (maintainer-confirmed 2026-09-05) was withdrawn 2026-09-07 — the updated https://nan.builders/docs/models still states '262K token context, the model's native window' and models.dev agrees at 262,144; NaN docs are treated as the most reliable source (maintainer instruction, 2026-09-07)."
 ],
 } as const;

package/src/commands.ts

@@ -1,45 +1,40 @@
 /**
- * `/nan-mcp` — slash command configuring BOTH MCP bridges this package
- * registers as pi tools:
+ * `/nan-mcp` — slash command configuring both native MCP servers this
+ * package registers:
  *
- * - `web-search` — the official NaN remote MCP server (api.nan.builders/mcp),
- *   exposing `nan_web_search`. Default: enabled.
- * - `nan-mcp-server` — the community stdio media server (image generation /
- *   editing, TTS, STT). Default: enabled, lazy (spawned per tool call).
+ * - `nan-search` — the official NaN remote MCP server (api.nan.builders/mcp),
+ *   exposing `mcp__nan-search__web_search`. Default: enabled.
+ * - `nan-media` — the community stdio media server (flux-2-klein / kokoro /
+ *   whisper), exposing `mcp__nan-media__generate_image` etc. Default: enabled.
  *
- * pi has no MCP client and therefore no `/mcp` command of its own, so this
- * command owns the toggles:
- *
- * - `/nan-mcp` or `/nan-mcp status` — state of both bridges and their source.
- * - `/nan-mcp enable [target]` — enable a bridge (or both, when no target is
- *   given) and persist it in `<agentDir>/nan-provider.json`; enabled bridges
- *   register their tools immediately for the current session.
- * - `/nan-mcp disable [target]` — disable persistently. pi has no
- *   unregisterTool, so already-registered tools remain until the next
- *   session/restart; the persisted toggle governs future sessions.
+ * Both are session-scoped native servers (visible in /mcp, source "extension").
+ * /nan-mcp enables/disables them via pi.registerMcpServer / pi.unregisterMcpServer,
+ * persisting the toggle in nan-provider.json.
  *
  * Targets accept aliases: `web-search` (`search`, `web_search`, `official`,
  * `nan-web-search`) and `nan-mcp-server` (`media`, `media-mcp`, `nan-media`).
  * Explicit env vars (NAN_MCP_TOOLS, NAN_MEDIA_MCP) override the persisted
  * toggles for the session.
+ *
+ * Native behavior changes from the old bridge:
+ * - `disable` IMMEDIATELY hides tools (unregisterMcpServer disconnects the
+ *   server; the MCP extension hides tools when disconnected).
+ * - `enable` IMMEDIATELY registers the server for the current session.
+ * - `status` reads the server state from pi.getMcpServers().
+ * - Missing API key at load: server is NOT registered; guidance notification
+ *   replaces the old "errors at call time" behavior.
  */
 
 import type { ExtensionAPI, ExtensionCommandContext } from "@earendil-works/pi-coding-agent";
-import { mediaMcpCommand, mediaMcpEnabled, mediaMcpSource } from "./mcp/nan-media.ts";
-import { webSearchBridgeEnabled, webSearchBridgeSource } from "./mcp/nan-search.ts";
-import { NAN_STATE_FILE, readBridgeState, writeBridgeState, type BridgeKey } from "./mcp/state.ts";
-
-export interface NanMcpCommandOptions {
-	/** Register web-search bridge tools now; must skip already-registered tools. */
-	registerWebSearchTools: () => void;
-	/** Register media bridge tools now; must skip already-registered tools. */
-	registerMediaTools: () => void;
-}
+import { mediaMcpCommand, mediaMcpEnabled, mediaMcpSource, mediaMcpEnvExplicit, mediaMcpEnvTruthy, mediaMcpTimeoutSec } from "./mcp/media-server.ts";
+import { webSearchBridgeEnabled, webSearchBridgeSource, mcpToolsEnvExplicit, mcpToolsDisabled } from "./mcp/api-key.ts";
+import { NAN_API_KEY_ENV, tryResolveNanApiKeyViaRegistry } from "./mcp/api-key.ts";
+import { NAN_STATE_FILE, readBridgeState, writeBridgeState, type BridgeKey, readState, writeState } from "./mcp/state.ts";
 
 const USAGE =
 	"Usage: /nan-mcp [status] · /nan-mcp enable [web-search|nan-mcp-server] · /nan-mcp disable [web-search|nan-mcp-server]";
 
-/** Canonical bridge targets with aliases (e.g. `/mcp enable nan-mcp-server` muscle memory). */
+/** Canonical bridge targets with aliases. */
 const TARGETS: Record<string, BridgeKey> = {
 	"web-search": "webSearch",
 	search: "webSearch",
@@ -67,19 +62,64 @@ function sourceLabel(bridge: BridgeKey, source: "env" | "persisted" | "default")
 	if (source === "persisted") {
 		return `persisted in <agentDir>/${NAN_STATE_FILE} (${targetName(bridge)}: ${readBridgeState(bridge)})`;
 	}
-	return "default (both bridges are enabled and lazy by default)";
+	return "default (both bridges are enabled by default)";
+}
+
+/** Resolve a bridge key to its native server name. */
+function serverName(bridge: BridgeKey): string {
+	return bridge === "webSearch" ? "nan-search" : "nan-media";
+}
+
+/** Describe the bridge in the status message. */
+function bridgeDescription(bridge: BridgeKey): string {
+	return bridge === "webSearch"
+		? "web-search bridge (official NaN MCP → mcp__nan-search__web_search)"
+		: "nan-mcp-server bridge (community media MCP → mcp__nan-media__generate_image/edit_image/text_to_speech/list_voices/speech_to_text)";
+}
+
+/** Check if a native server of this name is currently registered. */
+function nativeServerRegistered(pi: ExtensionAPI, name: string): boolean {
+	try {
+		return pi.getMcpServers().some((s) => s.name === name);
+	} catch {
+		return false;
+	}
 }
 
-function statusMessage(): string {
-	return [
-		`web-search bridge (official NaN MCP → nan_web_search): ${webSearchBridgeEnabled() ? "enabled" : "disabled"} — ${sourceLabel("webSearch", webSearchBridgeSource())}.`,
-		`nan-mcp-server bridge (community media MCP → nan_generate_image/nan_edit_image/nan_text_to_speech/nan_list_voices/nan_speech_to_text): ${mediaMcpEnabled() ? "enabled" : "disabled"} — ${sourceLabel("mediaMcp", mediaMcpSource())}.`,
-		`Media spawn command: ${mediaMcpCommand().join(" ")} (per tool call, terminated after).`,
-		"Configure with /nan-mcp enable|disable [web-search|nan-mcp-server]; no target = both.",
-	].join("\n");
+/** Status message listing both bridges' gate state and native registration. */
+function statusMessage(pi: ExtensionAPI): string {
+	const lines: string[] = [];
+
+	for (const bridge of ["webSearch" as BridgeKey, "mediaMcp" as BridgeKey]) {
+		const enabled = bridge === "webSearch" ? webSearchBridgeEnabled() : mediaMcpEnabled();
+		const source = bridge === "webSearch" ? webSearchBridgeSource() : mediaMcpSource();
+		const regStatus = nativeServerRegistered(pi, serverName(bridge)) ? "registered (native)" : "not registered";
+		lines.push(`${bridgeDescription(bridge)}: ${enabled ? "enabled" : "disabled"} — ${sourceLabel(bridge, source)}, ${regStatus}.`);
+
+		if (bridge === "webSearch") {
+			// Key status for web-search
+			const keyStatus = resolveKeyStatus(pi);
+			lines.push(`  API key: ${keyStatus}.`);
+		}
+		if (bridge === "mediaMcp") {
+			lines.push(`  Media spawn command: ${mediaMcpCommand().join(" ")}.`);
+		}
+	}
+
+	lines.push("Configure with /nan-mcp enable|disable [web-search|nan-mcp-server]; no target = both.");
+	return lines.join("\n");
+}
+
+/** Check API key resolution status for web-search. */
+async function resolveKeyStatus(pi: ExtensionAPI): Promise<string> {
+	// @ts-expect-error — modelRegistry is a runtime surface.
+	const registryKey = await tryResolveNanApiKeyViaRegistry(pi.modelRegistry);
+	const envKey = process.env[NAN_API_KEY_ENV];
+	if (registryKey || envKey) return "resolved";
+	return "NOT SET — run /login nan or export NAN_API_KEY";
 }
 
-/** Parse an optional target argument; no target = both bridges. Returns undefined target name on unknown. */
+/** Parse an optional target argument; no target = both bridges. */
 function parseTarget(token: string | undefined): { bridges: BridgeKey[]; target?: string } | undefined {
 	if (token === undefined) return { bridges: ["webSearch", "mediaMcp"] };
 	const key = TARGETS[token.toLowerCase()];
@@ -92,18 +132,25 @@ function describeBridges(bridges: BridgeKey[]): string {
 	return names.length === 2 ? "both bridges" : names.join(" and ");
 }
 
-export function registerNanMcpCommand(pi: ExtensionAPI, options: NanMcpCommandOptions): void {
+/**
+ * Check if a bridge is effectively enabled (gate resolved).
+ */
+function bridgeEffectivelyEnabled(bridge: BridgeKey): boolean {
+	return bridge === "webSearch" ? webSearchBridgeEnabled() : mediaMcpEnabled();
+}
+
+export function registerNanMcpCommand(pi: ExtensionAPI): void {
 	pi.registerCommand("nan-mcp", {
 		description:
-			"NaN MCP configuration: enable/disable/status for both bridges (official web-search + community nan-mcp-server)",
+			"NaN MCP configuration: enable/disable/status for both native servers (official web-search + community media)",
 		getArgumentCompletions: (argumentPrefix: string) => {
 			const prefix = argumentPrefix.trim().toLowerCase();
 			const items = [
-				{ value: "enable", label: "enable", description: "Enable a bridge (or both) and persist it" },
-				{ value: "disable", label: "disable", description: "Disable a bridge (or both) persistently" },
-				{ value: "status", label: "status", description: "Show current bridge status" },
-				{ value: "enable web-search", label: "enable web-search", description: "Enable the official NaN web-search bridge" },
-				{ value: "enable nan-mcp-server", label: "enable nan-mcp-server", description: "Enable the community media bridge" },
+				{ value: "enable", label: "enable", description: "Enable a native server (or both) and persist it" },
+				{ value: "disable", label: "disable", description: "Disable a native server (or both) persistently" },
+				{ value: "status", label: "status", description: "Show current server status and registration" },
+				{ value: "enable web-search", label: "enable web-search", description: "Enable the official NaN web-search server" },
+				{ value: "enable nan-mcp-server", label: "enable nan-mcp-server", description: "Enable the community media server" },
 			];
 			const filtered = items.filter((item) => item.value.startsWith(prefix));
 			return filtered.length > 0 ? filtered : null;
@@ -114,12 +161,11 @@ export function registerNanMcpCommand(pi: ExtensionAPI, options: NanMcpCommandOp
 			const subcommand = rawSubcommand.toLowerCase();
 
 			if (subcommand === "status") {
-				ctx.ui.notify(statusMessage(), "info");
+				ctx.ui.notify(statusMessage(pi), "info");
 				return;
 			}
 
 			if (subcommand === "enable" || subcommand === "disable") {
-				// Trailing tokens beyond the target (e.g. duplicate target names) are ignored.
 				const parsed = parseTarget(rawTarget);
 				if (parsed === undefined || rest.length > 0) {
 					ctx.ui.notify(
@@ -136,23 +182,57 @@ export function registerNanMcpCommand(pi: ExtensionAPI, options: NanMcpCommandOp
 				for (const bridge of bridges) {
 					writeBridgeState(bridge, enabled);
 				}
-				if (enabled) {
-					if (bridges.includes("webSearch")) options.registerWebSearchTools();
-					if (bridges.includes("mediaMcp")) options.registerMediaTools();
+
+				// Immediate native registration/unregistration.
+				for (const bridge of bridges) {
+					const name = serverName(bridge);
+					if (enabled) {
+						// Only register if gate allows it.
+						if (bridgeEffectivelyEnabled(bridge)) {
+							if (bridge === "webSearch") {
+								// @ts-expect-error — modelRegistry is a runtime surface.
+								const registryKey = await tryResolveNanApiKeyViaRegistry(pi.modelRegistry);
+								const apiKey = registryKey ?? process.env[NAN_API_KEY_ENV];
+								if (apiKey) {
+									pi.registerMcpServer(name, {
+										type: "http",
+										url: "https://api.nan.builders/mcp",
+										headers: { Authorization: `Bearer ${apiKey}` },
+										exposure: "direct",
+									});
+								} else {
+									ctx.ui.notify(
+										`${NAN_API_KEY_ENV} is not set. Export it or run /login nan.`,
+										"warning",
+									);
+								}
+							} else {
+								const mediaArgs = mediaMcpCommand();
+								pi.registerMcpServer(name, {
+									type: "stdio",
+									command: mediaArgs[0]!,
+									args: mediaArgs.slice(1),
+									env: { [NAN_API_KEY_ENV]: process.env[NAN_API_KEY_ENV] ?? "" },
+									exposure: "direct",
+									timeout: mediaMcpTimeoutSec(),
+								});
+							}
+						}
+					} else {
+						pi.unregisterMcpServer(name);
+					}
 				}
 
 				const where = target ? `for "${target}"` : "for both bridges";
 				const persistence = `persisted in <agentDir>/${NAN_STATE_FILE}`;
 				if (enabled) {
 					ctx.ui.notify(
-						`Enabled ${describeBridges(bridges)} ${where}, ${persistence}. Tools are available now; both bridges are lazy (web-search calls the endpoint per request, nan-mcp-server spawns per call), so nothing runs until invoked.` +
-							" An explicit NAN_MCP_TOOLS / NAN_MEDIA_MCP env var would override this toggle.",
+						`Enabled ${describeBridges(bridges)} ${where}, ${persistence}. Native servers registered for this session.`,
 						"info",
 					);
 				} else {
 					ctx.ui.notify(
-						`Disabled ${describeBridges(bridges)} ${where}, ${persistence}. pi has no unregisterTool, so tools already registered in this session remain until restart (or /reload); future sessions will not register them.` +
-							" An explicit NAN_MCP_TOOLS / NAN_MEDIA_MCP env var overrides this toggle.",
+						`Disabled ${describeBridges(bridges)} ${where}, ${persistence}. Native servers unregistered — tools hidden immediately.`,
 						"warning",
 					);
 				}

package/src/context-overflow-classifier.ts

@@ -1,7 +1,7 @@
 /**
  * NaN-aware context-overflow classification.
  *
- * Root cause (pi-ai, still present on 0.85.1 / main): when history is replayed
+ * Root cause (pi-ai, verified on 0.87.1): when history is replayed
  * into a DIFFERENT model, `transformMessages` downgrades every non-redacted
  * `thinking` block to plain `text` verbatim, with no size bound — and nothing
  * bounds the SUM across messages. Switching from a 1M-context model

package/src/cross-model-thinking-guard.ts

@@ -1,7 +1,7 @@
 /**
  * Cross-model thinking guard.
  *
- * Root cause this guards against (pi-ai, still present on 0.85.1 / main): when
+ * Root cause this guards against (pi-ai, verified on 0.87.1): when
  * history is replayed into a DIFFERENT model, `transformMessages` downgrades
  * every non-redacted `thinking` block to a plain `text` block verbatim
  * (`packages/ai/src/api/transform-messages.ts`), and `openai-completions`