@prom.codes/memory-mcp 0.10.0 → 0.10.2

package/dist/bin.js

@@ -194,9 +194,6 @@ function isHomeOrFilesystemRoot(root) {
   return false;
 }
 
-// ../shared/dist/index.js
-var PROMETHEUS_VERSION = "0.1.0";
-
 // dist/composition.js
 import { createHash } from "node:crypto";
 import { homedir as homedir3 } from "node:os";
@@ -876,6 +873,7 @@ var DEFAULT_BASE2 = "https://api.prom.codes";
 var DEFAULT_NAME = "prometheus";
 var DEFAULT_MODEL2 = "prometheus";
 var DEFAULT_BATCH4 = 100;
+var DEFAULT_BATCH_CHARS2 = 4e5;
 var DEFAULT_RETRIES4 = 4;
 var DEFAULT_BACKOFF4 = 250;
 function sleep4(ms, signal) {
@@ -907,6 +905,7 @@ var PrometheusRerankProvider = class {
   #apiKey;
   #url;
   #batchSize;
+  #maxBatchChars;
   #maxRetries;
   #retryBaseMs;
   #fetch;
@@ -924,6 +923,7 @@ var PrometheusRerankProvider = class {
     this.#apiKey = opts.apiKey;
     this.#url = `${(opts.baseUrl ?? DEFAULT_BASE2).replace(/\/+$/, "")}/rerank`;
     this.#batchSize = opts.batchSize ?? DEFAULT_BATCH4;
+    this.#maxBatchChars = opts.maxBatchChars ?? DEFAULT_BATCH_CHARS2;
     this.#maxRetries = opts.maxRetries ?? DEFAULT_RETRIES4;
     this.#retryBaseMs = opts.retryBaseMs ?? DEFAULT_BACKOFF4;
     this.#fetch = opts.fetch ?? fetch;
@@ -937,21 +937,44 @@ var PrometheusRerankProvider = class {
       return [];
     const all = new Array(candidates.length);
     let cursor = 0;
-    for (let start = 0; start < candidates.length; start += this.#batchSize) {
-      const slice = candidates.slice(start, start + this.#batchSize);
+    let start = 0;
+    while (start < candidates.length) {
+      const end = this.#batchEnd(query, candidates, start);
+      const slice = candidates.slice(start, end);
       const scored = await this.#rerankBatch(query, slice, opts?.signal);
       for (const hit of scored) {
         const globalIndex = start + hit.localIndex;
         const cand = candidates[globalIndex];
         all[cursor++] = { id: cand.id, index: globalIndex, score: hit.score };
       }
+      start = end;
     }
+    all.length = cursor;
     all.sort((a, b) => b.score - a.score);
     if (opts?.topK !== void 0 && opts.topK >= 0 && opts.topK < all.length) {
       return all.slice(0, opts.topK);
     }
     return all;
   }
+  /**
+   * Greedy batch cut honouring both the document-count cap and the
+   * char budget (query + documents must stay under the proxy limit).
+   * Always advances by at least one candidate, even if a single
+   * document alone exceeds the budget (the proxy applies upstream
+   * truncation; refusing to send it would strand the rest).
+   */
+  #batchEnd(query, candidates, start) {
+    let chars = query.length;
+    let end = start;
+    while (end < candidates.length && end - start < this.#batchSize) {
+      const len = candidates[end].text.length;
+      if (end > start && chars + len > this.#maxBatchChars)
+        break;
+      chars += len;
+      end += 1;
+    }
+    return end;
+  }
   async #rerankBatch(query, batch, signal) {
     const init = {
       method: "POST",
@@ -2907,14 +2930,16 @@ var setupInput = {
   runtimes: z.array(runtimeEnum).min(1).optional()
 };
 var emptyInput = {};
-function registerTools(server, deps) {
-  const { backend, workspaceRoot, projectId, projectName, dbPath, extractor } = deps;
-  const mirrorToFiles = !deps.rootIsHomeOrFsRoot;
+function registerTools(server, source) {
+  const ready = typeof source === "function" ? source : () => Promise.resolve(source);
   server.registerTool("read", {
     title: "Recall agent memory",
     description: "Read agent memory for this project along the scope chain (project \u2192 workspace \u2192 tenant \u2192 system; narrowest scope wins). Syncs `.prometheus/memories/*.md` first, then returns the resolved records plus a prompt-ready `woven` markdown block (token-capped). Call this at the START of a session or task to recall what earlier sessions learned.",
     inputSchema: readInput
   }, async (args) => {
+    const deps = await ready();
+    const { backend, workspaceRoot, projectId, projectName } = deps;
+    const mirrorToFiles = !deps.rootIsHomeOrFsRoot;
     const limit = clampLimit(args.limit, DEFAULT_READ_LIMIT);
     const synced = mirrorToFiles ? await syncProjectFiles(backend, { projectId, workspaceRoot }) : 0;
     const records = await backend.read({
@@ -2935,6 +2960,9 @@ function registerTools(server, deps) {
     description: "Upsert one memory record (identity: scope+type+key). Use type `semantic` for durable facts, `procedural` for how-to knowledge, `episodic` for session events, `working` for short-lived notes. Default scope `project` also mirrors the value to `.prometheus/memories/<key>.md` (git-versioned, human-editable). Values matching the secret deny-list are rejected. Call this whenever the user states a durable preference, decision, or correction worth remembering.",
     inputSchema: writeInput
   }, async (args) => {
+    const deps = await ready();
+    const { backend, workspaceRoot, projectId } = deps;
+    const mirrorToFiles = !deps.rootIsHomeOrFsRoot;
     assertValueSize(args.value);
     assertNoSecrets(`${args.key}
 ${args.value}`);
@@ -2962,6 +2990,7 @@ ${args.value}`);
     description: "Session-end consolidation: `plan`/`outcome` become one episodic record (key = sessionId), `facts` become semantic upserts, `procedures` become procedural upserts. Secret-bearing payloads are rejected. Call this at the END of a session to persist what was learned.",
     inputSchema: captureInput
   }, async (args) => {
+    const { backend, projectId, extractor } = await ready();
     const texts = [
       args.plan ?? "",
       args.outcome ?? "",
@@ -3025,6 +3054,9 @@ ${f.value}`);
     description: "Full-text search (FTS5) over memory keys and values within this project's scope chain, ranked by relevance. Returns matching records plus a highlighted snippet per hit. Use this when memory_read's recall is not specific enough. Does not bump useCount.",
     inputSchema: searchInput
   }, async (args) => {
+    const deps = await ready();
+    const { backend, workspaceRoot, projectId } = deps;
+    const mirrorToFiles = !deps.rootIsHomeOrFsRoot;
     const limit = clampLimit(args.limit, 20);
     if (mirrorToFiles)
       await syncProjectFiles(backend, { projectId, workspaceRoot });
@@ -3048,6 +3080,7 @@ ${f.value}`);
     description: "Flat listing of this project's memory records without scope resolution \u2014 inspection/debug surface. Optional filters: scope, type, keyContains (case-insensitive substring).",
     inputSchema: listInput
   }, async (args) => {
+    const { backend, projectId, projectName, dbPath } = await ready();
     const limit = clampLimit(args.limit, DEFAULT_READ_LIMIT);
     const records = await backend.list({
       projectId,
@@ -3068,6 +3101,9 @@ ${f.value}`);
     description: "Delete one memory record by identity (scope+type+key). For project-scoped semantic records the mirrored `.prometheus/memories/<key>.md` file is removed as well. Returns whether a record/file was actually removed.",
     inputSchema: deleteInput
   }, async (args) => {
+    const deps = await ready();
+    const { backend, workspaceRoot, projectId } = deps;
+    const mirrorToFiles = !deps.rootIsHomeOrFsRoot;
     const scope = args.scope;
     const removed = await backend.delete({
       projectId,
@@ -3087,6 +3123,9 @@ ${f.value}`);
     description: "Idempotently install the Prometheus memory-protocol rule block into agent runtime configs in this workspace: CLAUDE.md (claude-code), .cursor/rules/prometheus-memory.mdc (cursor), .augment/rules/prometheus-memory.md (augment), AGENTS.md (agents). Without `runtimes` it auto-detects which runtimes are present (fallback: agents). Only the marked block is written \u2014 existing content is never touched. Re-running updates the block in place.",
     inputSchema: setupInput
   }, async (args) => {
+    const deps = await ready();
+    const { workspaceRoot } = deps;
+    const mirrorToFiles = !deps.rootIsHomeOrFsRoot;
     if (!mirrorToFiles) {
       return textResult({
         workspaceRoot,
@@ -3106,6 +3145,9 @@ ${f.value}`);
     description: "Health check for this project's agent memory. Reports the resolved workspace root, project id, DB path, how many records are stored (total + by scope), the embedding provider with a zero-cost key-reachability probe, and which quality levers are active (rerank / rewrite / temporal). CALL THIS to confirm where memory is stored, how much is there, and whether the API key works.",
     inputSchema: emptyInput
   }, async () => {
+    const deps = await ready();
+    const { backend, workspaceRoot, projectId, projectName, dbPath } = deps;
+    const mirrorToFiles = !deps.rootIsHomeOrFsRoot;
     const stats = await backend.stats(projectId);
     let embeddingsReachable = null;
     let embeddingsError = null;
@@ -3153,7 +3195,7 @@ ${f.value}`);
 // dist/server.js
 var SERVER_IDENTITY = {
   name: "prometheus-memory-mcp",
-  version: PROMETHEUS_VERSION,
+  version: "0.10.2",
   title: "prom.codes Memory"
 };
 var SERVER_INSTRUCTIONS = "Persistent agent memory for this workspace \u2014 USE IT PROACTIVELY; the user will not tell you to. Protocol:\n1. ONE-TIME: if this workspace has no Prometheus memory rule yet, call memory_setup now (idempotent) so the protocol is installed into the runtime rule files and survives future sessions. (The server also auto-installs it on startup when a project rule file already exists \u2014 memory_setup covers the rest.)\n2. SESSION START: before any non-trivial task, call memory_read to recall facts, decisions and procedures from earlier sessions.\n3. DURING WORK: when the user states a durable preference, decision, correction or project fact, store it with memory_write (semantic for facts, procedural for how-tos) \u2014 without being asked.\n4. LOOK-UP: use memory_search for keyword recall when memory_read is not specific enough.\n5. SESSION END: consolidate what was learned with memory_capture.\nCall memory_status anytime to check what is stored and whether the rule is installed. Never store secrets, API keys or credentials \u2014 such writes are rejected.";
@@ -3175,6 +3217,11 @@ async function main() {
     instructions: SERVER_INSTRUCTIONS
   });
   let composed = null;
+  let composedResolve;
+  const composedReady = new Promise((res) => {
+    composedResolve = res;
+  });
+  registerTools(server, () => composedReady);
   const shutdown = async (signal) => {
     process.stderr.write(`prometheus-memory-mcp: received ${signal}, shutting down
 `);
@@ -3208,7 +3255,7 @@ async function main() {
       }).catch(() => {
       });
     }
-    registerTools(server, composed);
+    composedResolve(composed);
   };
   if (eagerVia !== null) {
     boot(void 0, eagerVia);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@prom.codes/memory-mcp",
-  "version": "0.10.0",
+  "version": "0.10.2",
   "description": "prom.codes Memory — persistent, local-first agent memory as an MCP server.",
   "type": "module",
   "bin": {