Expose missing-logprob probes in MCP

packages/choosekit-mcp/README.md

@@ -15,6 +15,7 @@ Select and configure the backend when the MCP process starts. `CHOOSEKIT_BACKEND
 | All | `CHOOSEKIT_MODE` | `labels` | `labels` for every backend; `minimal-prefix` is llama.cpp only |
 | All | `CHOOSEKIT_IMAGE_ROOT` | Current working directory | Root directory for files supplied through `imagePaths` |
 | llama.cpp | `CHOOSEKIT_BASE_URL` | Required | Base URL of the llama.cpp server, for example `http://127.0.0.1:8080` |
+| llama.cpp | `CHOOSEKIT_PROBE_MISSING_LOGPROBS` | `false` | Set to `true` to score choices outside the returned top logprobs with additional requests |
 | Ollama | `CHOOSEKIT_BASE_URL` | `http://127.0.0.1:11434` | Base URL of the Ollama server |
 | OpenRouter | `OPENROUTER_API_KEY` | Required | OpenRouter API key |
 | OpenRouter | `OPENROUTER_PROVIDER` | Not set | Pins one provider and disables fallback |
@@ -93,7 +94,7 @@ Relative paths are resolved from `CHOOSEKIT_IMAGE_ROOT`, which defaults to the M
 
 With `labels`, choosekit maps the supplied choice keys to A/B/C labels for scoring, so each description must contain the option's full meaning. With `minimal-prefix`, it scores the shortest token prefixes that distinguish the original keys. When no supplied option may apply, add an explicit choice such as `insufficient_information`.
 
-The result contains the selected key, the complete normalized distribution, raw scores, margin, entropy, token-boundary rollback, and backend usage when available. A score is `null` when the upstream backend did not return a log probability for that choice; its probability in the distribution is `0`. Probabilities represent relative preference among the supplied choices. Estimating correctness requires separate calibration.
+The result contains the selected key, the normalized distribution, raw scores, margin, entropy, token-boundary rollback, and backend usage when available. A score is `null` when the upstream backend did not return a log probability for that choice; its probability in the distribution is `0`. Probabilities represent relative preference among the supplied choices. Estimating correctness requires separate calibration.
 
 ## Requirements
 

packages/choosekit-mcp/package.json

@@ -1,6 +1,6 @@
 {
   "name": "choosekit-mcp",
-  "version": "0.3.0",
+  "version": "0.4.0",
   "description": "Use choosekit as an MCP server.",
   "license": "Apache-2.0",
   "author": {
@@ -37,7 +37,7 @@
   },
   "dependencies": {
     "@modelcontextprotocol/server": "2.0.0",
-    "choosekit": "^0.6.0",
+    "choosekit": "^0.7.0",
     "zod": "^4.6.4"
   },
   "devDependencies": {

packages/choosekit-mcp/src/cli.ts

@@ -18,6 +18,7 @@ try {
         baseURL: config.baseURL,
         ...(config.model === undefined ? {} : { model: config.model }),
         mode: config.mode,
+        probeMissingLogprobs: config.probeMissingLogprobs,
       });
       break;
     case "ollama":

packages/choosekit-mcp/src/config.ts

@@ -9,6 +9,7 @@ export type Config =
     readonly imageRoot?: string;
     readonly model?: string;
     readonly mode: "labels" | "minimal-prefix";
+    readonly probeMissingLogprobs: boolean;
   }
   | {
     readonly backend: "openrouter";
@@ -89,11 +90,16 @@ export function loadConfig(env: NodeJS.ProcessEnv = process.env): Config {
   if (mode !== "labels" && mode !== "minimal-prefix") {
     throw new TypeError("CHOOSEKIT_MODE must be labels or minimal-prefix.");
   }
+  const probe = env.CHOOSEKIT_PROBE_MISSING_LOGPROBS?.trim() ?? "false";
+  if (probe !== "true" && probe !== "false") {
+    throw new TypeError("CHOOSEKIT_PROBE_MISSING_LOGPROBS must be true or false.");
+  }
   return {
     backend,
     baseURL,
     ...(imageRoot === undefined ? {} : { imageRoot }),
     ...(model === undefined ? {} : { model }),
     mode,
+    probeMissingLogprobs: probe === "true",
   };
 }

packages/choosekit-mcp/tests/cli.test.mjs

@@ -25,6 +25,7 @@ const configurationKeys = [
   "CHOOSEKIT_IMAGE_ROOT",
   "CHOOSEKIT_MODEL",
   "CHOOSEKIT_MODE",
+  "CHOOSEKIT_PROBE_MISSING_LOGPROBS",
   "OPENROUTER_API_KEY",
   "OPENROUTER_PROVIDER",
 ];
@@ -134,6 +135,8 @@ test("reports invalid environment configuration without writing to stdout", () =
       "CHOOSEKIT_BASE_URL must be a valid URL."],
     ["invalid mode", { CHOOSEKIT_MODE: "keys" },
       "CHOOSEKIT_MODE must be labels or minimal-prefix."],
+    ["invalid missing-logprob setting", { CHOOSEKIT_PROBE_MISSING_LOGPROBS: "1" },
+      "CHOOSEKIT_PROBE_MISSING_LOGPROBS must be true or false."],
     ["unknown backend", { CHOOSEKIT_BACKEND: "other" },
       "CHOOSEKIT_BACKEND must be llama-cpp, ollama, or openrouter."],
     ["Ollama without model", {
@@ -296,6 +299,7 @@ test("serves a minimal-prefix choice using the configured llama.cpp endpoint", a
     CHOOSEKIT_BASE_URL: `http://127.0.0.1:${port}`,
     CHOOSEKIT_MODEL: "fixture-model",
     CHOOSEKIT_MODE: "minimal-prefix",
+    CHOOSEKIT_PROBE_MISSING_LOGPROBS: "true",
   }));
   t.after(() => process_.close());
   await initialize(process_);

packages/choosekit-mcp/tests/package.test.mjs

@@ -18,6 +18,7 @@ function llamaEnvironment() {
     "CHOOSEKIT_BASE_URL",
     "CHOOSEKIT_MODEL",
     "CHOOSEKIT_MODE",
+    "CHOOSEKIT_PROBE_MISSING_LOGPROBS",
     "OPENROUTER_API_KEY",
     "OPENROUTER_PROVIDER",
   ]) delete env[key];