@atrib/summarize 0.4.2 → 0.4.3

package/README.md

@@ -30,14 +30,22 @@ mcp__atrib-summarize__summarize({
 
 OpenAI-compatible HTTP. Defaults to NVIDIA NIM with `qwen/qwen3.5-397b-a17b`. Override via env or per-call `model` input:
 
-| Env var | Default |
-|---|---|
-| `ATRIB_SUMMARIZE_API_KEY` | (fallback to `NVIDIA_API_KEY` then `NVIDIA_NIM_API_KEY`) |
-| `ATRIB_SUMMARIZE_BASE_URL` | `https://integrate.api.nvidia.com/v1` |
-| `ATRIB_SUMMARIZE_MODEL` | `qwen/qwen3.5-397b-a17b` |
-| `ATRIB_SUMMARIZE_MAX_TOKENS` | `4000` |
-| `ATRIB_SUMMARIZE_TEMPERATURE` | `0.3` |
-| `ATRIB_SUMMARIZE_TIMEOUT_MS` | `120000` |
+| Env var                       | Default                               |
+| ----------------------------- | ------------------------------------- |
+| `ATRIB_SUMMARIZE_API_KEY`     | fallback to provider env/cache        |
+| `ATRIB_SUMMARIZE_BASE_URL`    | `https://integrate.api.nvidia.com/v1` |
+| `ATRIB_SUMMARIZE_MODEL`       | `qwen/qwen3.5-397b-a17b`              |
+| `ATRIB_SUMMARIZE_MAX_TOKENS`  | `4000`                                |
+| `ATRIB_SUMMARIZE_TEMPERATURE` | `0.3`                                 |
+| `ATRIB_SUMMARIZE_TIMEOUT_MS`  | `120000`                              |
+
+Provider env/cache fallback:
+
+| Provider URL contains      | Env var              | Cache file                            |
+| -------------------------- | -------------------- | ------------------------------------- |
+| `integrate.api.nvidia.com` | `NVIDIA_API_KEY`     | `~/.atrib/secrets/nvidia-api-key`     |
+| `api.cerebras.ai`          | `CEREBRAS_API_KEY`   | `~/.atrib/secrets/cerebras-api-key`   |
+| `cloudflare.com`           | `CLOUDFLARE_API_KEY` | `~/.atrib/secrets/cloudflare-api-key` |
 
 Without an API key, the tool returns a warnings-only response per the [§5.8](https://github.com/creatornader/atrib/blob/main/atrib-spec.md#58-degradation-contract) graceful-degradation contract.
 
@@ -66,12 +74,15 @@ Add to your MCP host config:
     "command": "node",
     "args": ["/path/to/atrib-summarize/dist/main.js"],
     "env": {
-      "NVIDIA_API_KEY": "..."
+      "ATRIB_SUMMARIZE_MODEL": "qwen/qwen3.5-397b-a17b"
     }
   }
 }
 ```
 
+The API key can live in the host env or in the cache file above. Do not
+write secret values into shared MCP config.
+
 ## Status
 
 Initial scaffold (v0.1.0). 6 unit tests covering record selection (by hash, by context, unioned, missing-skip) + degradation paths (no inputs, no API key). Integration test against a real LLM is gated behind `ATRIB_SUMMARIZE_API_KEY` and not run in CI.

package/dist/index.js

@@ -75,7 +75,7 @@ async function handleSummarize(input) {
     }
     const llmCfg = resolveLlmConfig(input.model);
     if (!llmCfg) {
-        warnings.push('no LLM API key resolved (set ATRIB_SUMMARIZE_API_KEY or NVIDIA_API_KEY); cannot synthesize');
+        warnings.push('no LLM API key resolved (set ATRIB_SUMMARIZE_API_KEY, NVIDIA_API_KEY, or ~/.atrib/secrets/nvidia-api-key); cannot synthesize');
         return emptyOutput(warnings);
     }
     // Load local mirror once; filter to the requested set.

package/dist/llm.js

@@ -1,17 +1,59 @@
 // SPDX-License-Identifier: Apache-2.0
+import { existsSync, readFileSync } from 'node:fs';
+import { homedir } from 'node:os';
+import { join } from 'node:path';
+function readCacheSecret(cacheName) {
+    const home = process.env['HOME'] ?? homedir();
+    const path = join(home, '.atrib', 'secrets', cacheName);
+    if (!existsSync(path))
+        return '';
+    try {
+        return readFileSync(path, 'utf8').trim();
+    }
+    catch {
+        return '';
+    }
+}
+function firstSecret(...values) {
+    for (const value of values) {
+        const trimmed = value?.trim();
+        if (trimmed)
+            return trimmed;
+    }
+    return '';
+}
+function hostnameFor(baseUrl) {
+    try {
+        return new URL(baseUrl).hostname.toLowerCase();
+    }
+    catch {
+        return '';
+    }
+}
+function hostnameMatches(hostname, domain) {
+    return hostname === domain || hostname.endsWith(`.${domain}`);
+}
+function providerApiKey(baseUrl) {
+    const hostname = hostnameFor(baseUrl);
+    if (hostnameMatches(hostname, 'cerebras.ai')) {
+        return firstSecret(process.env['CEREBRAS_API_KEY'], readCacheSecret('cerebras-api-key'));
+    }
+    if (hostnameMatches(hostname, 'cloudflare.com')) {
+        return firstSecret(process.env['CLOUDFLARE_API_KEY'], readCacheSecret('cloudflare-api-key'));
+    }
+    return firstSecret(process.env['NVIDIA_API_KEY'], readCacheSecret('nvidia-api-key'), process.env['NVIDIA_NIM_API_KEY']);
+}
 /**
  * Resolve LLM config from env with documented defaults. Caller-supplied
  * model override (from the MCP tool input) wins over env.
  */
 export function resolveLlmConfig(modelOverride) {
-    const apiKey = process.env['ATRIB_SUMMARIZE_API_KEY'] ??
-        process.env['NVIDIA_API_KEY'] ??
-        process.env['NVIDIA_NIM_API_KEY'] ??
-        '';
+    const baseUrl = process.env['ATRIB_SUMMARIZE_BASE_URL'] ?? 'https://integrate.api.nvidia.com/v1';
+    const apiKey = firstSecret(process.env['ATRIB_SUMMARIZE_API_KEY'], providerApiKey(baseUrl));
     if (!apiKey)
         return null;
     return {
-        baseUrl: process.env['ATRIB_SUMMARIZE_BASE_URL'] ?? 'https://integrate.api.nvidia.com/v1',
+        baseUrl,
         model: modelOverride ?? process.env['ATRIB_SUMMARIZE_MODEL'] ?? 'qwen/qwen3.5-397b-a17b',
         apiKey,
         maxTokens: Number(process.env['ATRIB_SUMMARIZE_MAX_TOKENS'] ?? 4000),
@@ -50,7 +92,8 @@ export async function callLlm(cfg, systemMsg, userMsg) {
             throw new Error(`LLM POST ${res.status}: ${text.slice(0, 500)}`);
         }
         const json = (await res.json());
-        const content = json.choices?.[0]?.message?.content ?? '';
+        const message = json.choices?.[0]?.message;
+        const content = message?.content || message?.reasoning_content || '';
         if (!content)
             throw new Error('LLM response had empty content');
         return { content, model: json.model ?? cfg.model };

package/dist/prompt.js

@@ -64,6 +64,8 @@ function renderRecord(r) {
             out.push(`tool: ${sc.toolName}`);
         const c = sc.content;
         if (c) {
+            if (typeof c['tool_name'] === 'string')
+                out.push(`tool: ${c['tool_name']}`);
             if (typeof c['what'] === 'string')
                 out.push(`what: ${c['what']}`);
             if (typeof c['why_noted'] === 'string')
@@ -78,15 +80,19 @@ function renderRecord(r) {
                 out.push(`new_position: ${c['new_position']}`);
             if (typeof c['reason'] === 'string')
                 out.push(`reason: ${c['reason']}`);
+            if (c['args'])
+                out.push(`args (truncated): ${JSON.stringify(c['args']).slice(0, 1200)}`);
+            if (c['result'])
+                out.push(`result (truncated): ${JSON.stringify(c['result']).slice(0, 1200)}`);
             if (Array.isArray(c['topics'])) {
                 out.push(`topics: ${c['topics'].filter((x) => typeof x === 'string').join(', ')}`);
             }
         }
         // Tool-call sidecar may have args/result; keep brief to control prompt size.
         if (sc.args)
-            out.push(`args (truncated): ${JSON.stringify(sc.args).slice(0, 300)}`);
+            out.push(`args (truncated): ${JSON.stringify(sc.args).slice(0, 1200)}`);
         if (sc.result)
-            out.push(`result (truncated): ${JSON.stringify(sc.result).slice(0, 300)}`);
+            out.push(`result (truncated): ${JSON.stringify(sc.result).slice(0, 1200)}`);
     }
     else {
         out.push('(no semantic sidecar available, record predates local-mirror sidecar pattern; only cryptographic metadata is present)');

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@atrib/summarize",
-  "version": "0.4.2",
+  "version": "0.4.3",
   "description": "MCP server for atrib. Synthesizes a narrative across N records via an OpenAI-compatible LLM so agents read context, not raw record bytes.",
   "license": "Apache-2.0",
   "type": "module",
@@ -11,13 +11,13 @@
   "dependencies": {
     "@modelcontextprotocol/sdk": "^1.29.0",
     "zod": "^3.25.76",
-    "@atrib/mcp": "0.11.1"
+    "@atrib/mcp": "0.12.0"
   },
   "devDependencies": {
-    "@types/node": "^25.8.0",
-    "tsx": "^4.22.1",
+    "@types/node": "^25.9.1",
+    "tsx": "^4.22.3",
     "typescript": "^6.0.3",
-    "vitest": "^4.1.6"
+    "vitest": "^4.1.7"
   },
   "files": [
     "dist"