@bli-cockpit/mcp 0.1.21 → 0.1.22

package/README.md

@@ -34,7 +34,7 @@ about what an agent can reach.
 
 <!-- BEGIN GENERATED verb census — `npm run mcp:readme` -->
 
-**74 of 77 Tower verbs have an MCP twin.**
+**76 of 79 Tower verbs have an MCP twin.**
 Each tool goes through the SAME door its CLI verb calls, with the same
 collector device token — never a second route and never a service-role
 reader. `src/verb-census.test.ts` fails when a verb is in none of the
@@ -81,6 +81,8 @@ three tables below.
 | `cockpit memory log` | `memory_experience` | `POST /api/memory/experience` |
 | `cockpit model set` | `model_set` | `POST /api/settings/jarvis-model` |
 | `cockpit model show` | `model_show` | `GET /api/settings/jarvis-model` |
+| `cockpit models list` | `models_list` | `GET /api/models/cards` |
+| `cockpit models show` | `models_card` | `GET /api/models/cards?id=` |
 | `cockpit msg channels` | `msg_channels` | `GET /api/msg/channels` |
 | `cockpit msg create` | `msg_create_channel` | `POST /api/msg/channels` |
 | `cockpit msg dm` | `msg_dm` | `POST /api/msg/channels (dm)` |

package/dist/models-tools.d.ts

@@ -0,0 +1,25 @@
+/**
+ * `models_list` / `models_card` (BLI-3912) — the model-card shelf, for an
+ * agent, over the same `/api/models/cards` door `cockpit models` calls.
+ *
+ * These are the twins of `cockpit models list` and `cockpit models show`. They
+ * exist because "which model should I use for this, and what will it refuse"
+ * is a question an agent asks more often than a person does: the card carries
+ * the documented window, price, reasoning modes and tool support beside our own
+ * measured latency and probe verdicts, and a `stale:<field>` mark wherever the
+ * two disagree.
+ *
+ * The LINES come from the server, so the sentence an agent reads is the sentence
+ * a person reads in a terminal. Nothing is computed here beyond counting.
+ *
+ * Not to be confused with `model_show` / `model_set` next door, which are the
+ * ORG SETTING for which model a slot runs on. This is the shelf that setting
+ * picks from, and there is deliberately no write here: a card is changed by
+ * re-reading the provider (`npm run models:card add <id> --refresh`), never by
+ * an agent editing a number.
+ */
+import { type ToolDeps } from "./tool-result.js";
+export type ModelsDeps = ToolDeps;
+export declare function registerModelsTools(server: {
+    registerTool: (...args: never[]) => unknown;
+}, deps: ModelsDeps): void;

package/dist/models-tools.js

@@ -0,0 +1,73 @@
+/**
+ * `models_list` / `models_card` (BLI-3912) — the model-card shelf, for an
+ * agent, over the same `/api/models/cards` door `cockpit models` calls.
+ *
+ * These are the twins of `cockpit models list` and `cockpit models show`. They
+ * exist because "which model should I use for this, and what will it refuse"
+ * is a question an agent asks more often than a person does: the card carries
+ * the documented window, price, reasoning modes and tool support beside our own
+ * measured latency and probe verdicts, and a `stale:<field>` mark wherever the
+ * two disagree.
+ *
+ * The LINES come from the server, so the sentence an agent reads is the sentence
+ * a person reads in a terminal. Nothing is computed here beyond counting.
+ *
+ * Not to be confused with `model_show` / `model_set` next door, which are the
+ * ORG SETTING for which model a slot runs on. This is the shelf that setting
+ * picks from, and there is deliberately no write here: a card is changed by
+ * re-reading the provider (`npm run models:card add <id> --refresh`), never by
+ * an agent editing a number.
+ */
+import { z } from "zod";
+import { callAgentDoor } from "./agent-door.js";
+import { doorFailureText, errorResult, registrarFor, textResult, queryString, withSession, } from "./tool-result.js";
+export function registerModelsTools(server, deps) {
+    const register = registrarFor(server);
+    register("models_list", {
+        title: "List the model cards",
+        description: "Every model Tower knows about: the provider's documented context window, price per million, reasoning "
+            + "modes and tool support, beside our measured latency, observed tokens/s and last probe verdict. "
+            + "`STALE:<field>` means a measurement contradicted the documentation, with the command that re-reads the "
+            + "provider. Read-only.",
+        inputSchema: {},
+    }, async () => withSession(deps, async (session) => {
+        const response = await callAgentDoor(session, deps.fetchImpl, "GET", "/api/models/cards");
+        if (!response.ok)
+            return errorResult(doorFailureText("models_list", response));
+        const body = response.body;
+        const cards = (Array.isArray(body.cards) ? body.cards : []);
+        const lines = (Array.isArray(body.lines) ? body.lines : []);
+        const stale = cards.filter((card) => (card.stale?.length ?? 0) > 0).map((card) => card.id ?? "?");
+        const unreadable = (Array.isArray(body.unreadable) ? body.unreadable : []);
+        return textResult(`${cards.length} model card(s)`
+            + `${stale.length > 0 ? `, ${stale.length} stale: ${stale.join(", ")}` : ", none stale"}`
+            + `${unreadable.length > 0 ? `, ${unreadable.length} unreadable: ${unreadable.join(", ")}` : ""}`
+            + `${lines.length > 0 ? `\n${lines.join("\n")}` : ""}`, { cards, stale, unreadable });
+    }));
+    register("models_card", {
+        title: "Read one model card",
+        description: "One model in full: every documented value with the provider page it was read from, the quirks quoted "
+            + "verbatim, our measured latency and tokens/s, each probe verdict, and any stale mark. A field the "
+            + "provider does not state reads unknown with the reason — never a guess.",
+        inputSchema: {
+            id: z
+                .string()
+                .min(1)
+                .max(200)
+                .describe("The card id, `provider:model` (e.g. openai:gpt-5.6-terra), or one of its aliases."),
+        },
+    }, async (args) => withSession(deps, async (session) => {
+        const params = new URLSearchParams();
+        params.set("id", String(args.id));
+        const response = await callAgentDoor(session, deps.fetchImpl, "GET", `/api/models/cards${queryString(params)}`);
+        // A 404 here is the door's own sentence naming the id and the command
+        // that would add it, which is more useful than anything this file
+        // could compose.
+        if (!response.ok)
+            return errorResult(doorFailureText("models_card", response));
+        const body = response.body;
+        const card = (body.card ?? {});
+        const lines = (Array.isArray(body.lines) ? body.lines : []);
+        return textResult(lines.join("\n") || `Model card ${card.id ?? args.id}.`, { card });
+    }));
+}

package/dist/server.js

@@ -15,6 +15,7 @@ import { registerMailTools } from "./mail-tools.js";
 import { registerNotesTools } from "./notes-tools.js";
 import { registerNotesWriteTools } from "./notes-write-tools.js";
 import { registerMemoryExperienceTools } from "./memory-experience-tools.js";
+import { registerModelsTools } from "./models-tools.js";
 import { registerOpsTools } from "./ops-tools.js";
 import { registerPagesTools } from "./pages-tools.js";
 import { registerSearchTool } from "./search-tool.js";
@@ -404,6 +405,10 @@ export function createServer(deps) {
     // as the three families above, over the doors `cockpit jarvis` already
     // calls. It is the last Tower surface that had a CLI door and no MCP one.
     registerJarvisTools(server, { fetchImpl: deps.fetchImpl });
+    // BLI-3912: `models_*` — the model-card shelf, over the same door
+    // `cockpit models` calls. Read-only: a card is changed by re-reading the
+    // provider, never by an agent editing a number.
+    registerModelsTools(server, { fetchImpl: deps.fetchImpl });
     // BLI-3756 batch 1: the READS the CLI already had and this server did not —
     // the daily page, the meeting-notes library, the ops board, Slack coverage,
     // settings/team/model, Scout and the workbook. Same doors, same device

package/dist/verb-census.js

@@ -237,6 +237,10 @@ export const MCP_TWINS = {
     "team device-revoke": { tool: "team_device_revoke", door: "POST /api/ambient/devices/[deviceId]/revoke" },
     "model show": { tool: "model_show", door: "GET /api/settings/jarvis-model" },
     "model set": { tool: "model_set", door: "POST /api/settings/jarvis-model" },
+    // BLI-3912: the model-card shelf. `models` is the shelf; `model` next door is
+    // the org setting that picks from it.
+    "models list": { tool: "models_list", door: "GET /api/models/cards" },
+    "models show": { tool: "models_card", door: "GET /api/models/cards?id=" },
     "scout board": { tool: "scout_board", door: "GET /api/cockpit/scout" },
     "scout start": { tool: "scout_start", door: "POST /api/cockpit/scout (start)" },
     "scout dismiss": { tool: "scout_dismiss", door: "POST /api/cockpit/scout (dismiss)" },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bli-cockpit/mcp",
-  "version": "0.1.21",
+  "version": "0.1.22",
   "private": false,
   "description": "bli-tower \u2014 an MCP server over BLI Cockpit's agent doors: JARVIS (jarvis_*), documents (docs_*), channels (msg_*), issues (work_*), the daily page (brief_*), meeting notes (notes_*), the ops board (ops_status/slack_*), settings/team/model, Scout and the workbook, plus the legacy event-stream tools (emit_event, get_ticket_timeline, get_active_tickets).",
   "type": "module",