@miraigent/free-ai-ops-mcp 0.1.7 → 0.1.10

package/CHANGELOG.md

@@ -1,5 +1,31 @@
 # Changelog
 
+## [0.1.10] - 2026-06-14
+
+### Changed
+
+- Added a README tool picker so npm/GitHub visitors can choose the first MCP
+  tool by workflow risk before opening issues or integrating the package.
+
+## [0.1.9] - 2026-06-13
+
+### Added
+
+- Added a copy-ready JSON-RPC MCP session example so developers can verify
+  initialize, tools/list, and a human_review_gate call before integrating with a
+  desktop MCP client.
+- Added README guidance for using the session example as a public-safe smoke
+  test when opening GitHub issues.
+
+## [0.1.8] - 2026-06-12
+
+### Added
+
+- Added a copy-ready prompt_risk_review example for developers checking
+  customer-facing prompts before AI use.
+- Added README input/output for the prompt risk workflow and a matching npm
+  example script.
+
 ## [0.1.7] - 2026-06-10
 
 ### Added

package/README.md

@@ -16,6 +16,9 @@ Run the public MCP server:
 Then send a JSON-RPC tool call such as the copy-ready example in
 examples/human-review-gate/.
 
+For a complete public-safe MCP session that covers initialize, tools/list, and
+one tool call, see examples/mcp-json-rpc-session/.
+
 If you are not a developer and only want the spreadsheet/checklist version
 first, download the free review kit:
 
@@ -51,6 +54,20 @@ Claude Desktop config location:
 - Teams prototyping human-in-the-loop AI workflows before integrating real data.
 - Operators who can run npm/npx and want a small public MCP example.
 
+## Pick Your First Tool
+
+Start with the tool that matches the first unsafe handoff in your workflow:
+
+- AI draft may reach a customer: use `human_review_gate`.
+- Repeated support questions should become a public FAQ: use
+  `faq_candidate_review`.
+- CRM notes mix facts, AI suggestions, and human decisions: use
+  `ai_safe_crm_note`.
+- A prompt may touch customer data or public output: use `prompt_risk_review`.
+
+All examples use synthetic data. Keep private customer records, secrets,
+internal policy text, and paid product files out of public issues and screenshots.
+
 ## What To Do Next
 
 - Try one tool with synthetic data.
@@ -114,6 +131,52 @@ Use this as a small public proof before building a larger AI support workflow.
 See examples/human-review-gate/ for copy-ready JSON-RPC examples and a sample
 decision log that match the free Gumroad kit.
 
+## Copy-Ready MCP Session
+
+Use this when you want to confirm the package responds like an MCP server before
+connecting it to Claude Desktop, Cursor, or your own MCP client:
+
+    npm run mcp < examples/mcp-json-rpc-session/sample-session.jsonl
+
+The session sends three JSON-RPC messages:
+
+- initialize: confirms protocolVersion, capabilities, and serverInfo.
+- tools/list: confirms the four public alpha tools are visible.
+- tools/call: runs human_review_gate with synthetic support-reply data.
+
+If you open a GitHub issue, copy only synthetic input shapes and returned status
+fields. Do not paste private customer records, secrets, tokens, internal policy
+text, or MIRAI Memory behavior.
+
+## Check A Prompt Before AI Use
+
+Use prompt_risk_review before a support, CRM, FAQ, or workflow prompt is sent to
+AI. It helps developers decide whether a prompt can proceed with a checklist,
+needs human review, or should stop before use because it touches sensitive data
+or customer-facing output.
+
+Example input:
+
+    {
+      "operation": "support automation",
+      "promptSummary": "Draft a customer reply from inquiry details and suggest the next support step.",
+      "dataTypes": ["customer email", "inquiry body", "plan name"],
+      "customerFacing": true,
+      "riskLevel": "medium"
+    }
+
+Expected result:
+
+    {
+      "recommendation": "human_review_required",
+      "riskFlags": ["customer_facing", "sensitive_data_possible"],
+      "boundary": "This tool is a prompt risk helper. It is not legal advice and does not call an AI API."
+    }
+
+Run the copy-ready example:
+
+    npm run example:prompt-risk-review
+
 ## Launch Flow
 
 See LAUNCH_FLOW.md for the first public posting plan, issue collection flow, and
@@ -140,6 +203,10 @@ Run the public FREE-004 example:
 
     npm run example:human-review-gate
 
+Run the prompt risk review example:
+
+    npm run example:prompt-risk-review
+
 ## npm Package
 
 Package name: @miraigent/free-ai-ops-mcp

package/examples/mcp-json-rpc-session/README.md

@@ -0,0 +1,35 @@
+# MCP JSON-RPC Session Example
+
+This example gives developers a copy-ready MCP session for the public
+@miraigent/free-ai-ops-mcp package.
+
+Use it before wiring the package into Claude Desktop, Cursor, or a custom MCP
+client. It verifies that the server responds to:
+
+- initialize
+- tools/list
+- tools/call for human_review_gate
+
+## Run
+
+From the repository root:
+
+    npm run mcp < examples/mcp-json-rpc-session/sample-session.jsonl
+
+You should receive three JSON-RPC response lines. The final response should
+include a human_review_gate result with:
+
+- gateStatus: stop
+- reviewOwner: support lead
+- boundary: this tool is a review helper and does not send messages
+
+## Public-Safe Feedback
+
+If the output is unclear, open a Tried It feedback issue and include only:
+
+- which command you ran
+- the public-safe input shape
+- the returned status or field that was unclear
+
+Do not include secrets, credentials, private customer records, internal policy
+text, or MIRAI Memory behavior.

package/examples/mcp-json-rpc-session/sample-session.jsonl

@@ -0,0 +1,3 @@
+{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}
+{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}
+{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"human_review_gate","arguments":{"draftType":"customer support reply","audience":"customer","riskFlags":["refund","complaint","personal data"],"reviewOwner":"support lead","sendMode":"manual"}}}

package/examples/prompt-risk-review/README.md

@@ -0,0 +1,33 @@
+# Prompt Risk Review Example
+
+This example shows how to call the public prompt_risk_review MCP tool before an
+operational prompt is sent to AI.
+
+Use it when a support, CRM, FAQ, or workflow automation prompt might include
+customer-facing output or sensitive data.
+
+## Run
+
+From the repository root:
+
+    npm run example:prompt-risk-review
+
+The example sends sample-request.jsonl to the MCP server and prints the
+prompt_risk_review result.
+
+## What To Look For
+
+The sample includes a customer-facing support operation and customer email data,
+so the tool should return:
+
+- recommendation: human_review_required or stop_before_ai_use
+- riskFlags: customer_facing and sensitive_data_possible
+- saferNextStep: a human-reviewed data-handling step before real use
+
+## Safe Use
+
+Use synthetic examples only. Do not paste private customer records, secrets,
+contracts, payment details, or internal policy text into public issues,
+examples, or screenshots.
+
+This is an operations review helper, not legal or compliance advice.

package/examples/prompt-risk-review/run-example.mjs

@@ -0,0 +1,28 @@
+import { spawn } from 'node:child_process';
+import fs from 'node:fs';
+
+const child = spawn(process.execPath, ['mcp/free-ai-ops-server.mjs'], {
+  stdio: ['pipe', 'pipe', 'inherit']
+});
+
+const request = fs.readFileSync('examples/prompt-risk-review/sample-request.jsonl', 'utf8');
+let stdout = '';
+
+child.stdout.on('data', (chunk) => {
+  stdout += chunk;
+});
+
+child.stdin.write(request.trim() + '\n');
+child.stdin.end();
+
+await new Promise((resolve, reject) => {
+  child.on('error', reject);
+  child.on('close', (code) => {
+    if (code !== 0) reject(new Error('server exited with ' + code));
+    else resolve();
+  });
+});
+
+const response = JSON.parse(stdout.trim());
+const payload = JSON.parse(response.result.content[0].text);
+console.log(JSON.stringify(payload, null, 2));

package/examples/prompt-risk-review/sample-request.jsonl

@@ -0,0 +1 @@
+{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"prompt_risk_review","arguments":{"operation":"support automation","promptSummary":"Draft a customer reply from inquiry details and suggest the next support step.","dataTypes":["customer email","inquiry body","plan name"],"customerFacing":true,"riskLevel":"medium"}}}

package/mcp/free-ai-ops-server.mjs

@@ -1,6 +1,6 @@
 #!/usr/bin/env node
 
-const serverInfo = { name: 'miraigent-free-ai-ops-mcp', version: '0.1.7' };
+const serverInfo = { name: 'miraigent-free-ai-ops-mcp', version: '0.1.10' };
 
 const tools = [
   {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@miraigent/free-ai-ops-mcp",
-  "version": "0.1.7",
+  "version": "0.1.10",
   "description": "Free MCP server for developers adding human review gates, prompt risk checks, FAQ review, and CRM note safety to AI tools.",
   "homepage": "https://github.com/Miraigent/miraigent-free-ai-ops-mcp",
   "type": "module",
@@ -24,6 +24,7 @@
     "check": "node --check mcp/free-ai-ops-server.mjs && node --check scripts/smoke-test.mjs && node --check scripts/secret-scan.mjs && node scripts/secret-scan.mjs",
     "test": "node scripts/smoke-test.mjs",
     "example:human-review-gate": "node examples/human-review-gate/run-example.mjs",
+    "example:prompt-risk-review": "node examples/prompt-risk-review/run-example.mjs",
     "mcp": "node mcp/free-ai-ops-server.mjs"
   },
   "keywords": [