@vettly/mcp 0.1.7 → 0.1.9

package/README.md

@@ -1,17 +1,17 @@
 # @vettly/mcp
 
-MCP server for App Store Guideline 1.2 compliance. Enables AI assistants to make policy-governed, auditable content decisions.
+MCP server for UGC moderation. Enables AI assistants to make policy-governed, auditable content decisions.
 
-## App Store Guideline 1.2
+## UGC Moderation Essentials
 
-Apple requires every iOS app with user-generated content to implement four things. The MCP server lets AI assistants participate in all four:
+Apps with user-generated content need four things to stay compliant and keep users safe. The MCP server lets AI assistants participate in all four:
 
-| Requirement | Guideline | MCP Integration |
-|-------------|-----------|-----------------|
-| **Content filtering** | 1.2.1 | `moderate_content` tool |
-| **User reporting** | 1.2.2 | `get_recent_decisions` tool (review flagged content) |
-| **User blocking** | 1.2.3 | Pair with server-side SDK (`POST /v1/blocks`) |
-| **Audit trail** | — | Every tool response includes `decisionId` |
+| Requirement | MCP Integration |
+|-------------|-----------------|
+| **Content filtering** | `moderate_content` tool |
+| **User reporting** | `get_recent_decisions` tool (review flagged content) |
+| **User blocking** | Pair with server-side SDK (`POST /v1/blocks`) |
+| **Audit trail** | Every tool response includes `decisionId` |
 
 ```
 Prompt: "Check if this user comment is safe for our app: 'Great product!'
@@ -53,7 +53,7 @@ Add to your `claude_desktop_config.json`:
       "command": "npx",
       "args": ["-y", "@vettly/mcp"],
       "env": {
-        "VETTLY_API_KEY": "sk_live_xxx"
+        "VETTLY_API_KEY": "vettly_live_xxx"
       }
     }
   }

package/dist/bin/vettly-mcp.cjs

@@ -24,6 +24,13 @@ function registerModerateTool(server, client) {
       policyId: ModerateInputSchema.shape.policyId,
       contentType: ModerateInputSchema.shape.contentType
     },
+    {
+      title: "Moderate Content",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = ModerateInputSchema.parse(args);
       try {
@@ -80,6 +87,13 @@ function registerValidateTool(server, client) {
     {
       yamlContent: ValidateInputSchema.shape.yamlContent
     },
+    {
+      title: "Validate Policy",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = ValidateInputSchema.parse(args);
       try {
@@ -113,6 +127,13 @@ function registerListPoliciesTool(server, client) {
     "list_policies",
     "List all moderation policies available in your Vettly account.",
     {},
+    {
+      title: "List Policies",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async () => {
       try {
         const result = await client.listPolicies();
@@ -151,6 +172,13 @@ function registerUsageStatsTool(server, client) {
     {
       days: UsageStatsInputSchema.shape.days
     },
+    {
+      title: "Get Usage Stats",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = UsageStatsInputSchema.parse(args);
       try {
@@ -196,6 +224,13 @@ function registerDecisionsTool(server, client) {
       policyId: DecisionsInputSchema.shape.policyId,
       contentType: DecisionsInputSchema.shape.contentType
     },
+    {
+      title: "Get Recent Decisions",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = DecisionsInputSchema.parse(args);
       try {
@@ -316,6 +351,182 @@ function registerResources(server, client) {
   registerPolicyResources(server, client);
 }
 
+// src/prompts/moderate-content-guide.ts
+function registerModerateContentGuide(server) {
+  server.prompt(
+    "moderate-content-guide",
+    "Step-by-step guide to moderating content with Vettly. Explains the workflow: list policies, moderate content, and interpret results.",
+    async () => ({
+      messages: [
+        {
+          role: "user",
+          content: {
+            type: "text",
+            text: `You are helping a user moderate content with Vettly. Follow this workflow:
+
+## Step 1: List available policies
+Use the \`list_policies\` tool to see what moderation policies are configured. Each policy has an ID and defines rules for what content is allowed or flagged.
+
+## Step 2: Moderate content
+Use the \`moderate_content\` tool with:
+- \`content\`: The text, image URL, or video URL to check
+- \`policyId\`: The policy ID from step 1
+- \`contentType\`: "text", "image", or "video"
+
+## Step 3: Interpret results
+The response includes:
+- \`safe\`: Boolean \u2014 whether the content passed moderation
+- \`flagged\`: Boolean \u2014 whether the content was flagged for review
+- \`action\`: The recommended action ("allow", "flag", or "block")
+- \`categories\`: Breakdown of category scores (e.g. hate speech, violence, sexual content)
+- \`decisionId\`: Unique ID for this decision, useful for auditing
+
+## Tips
+- Always list policies first so you know which policy ID to use.
+- For images and video, pass the URL as the \`content\` parameter.
+- Use \`get_recent_decisions\` to review past moderation outcomes.
+- Use \`get_usage_stats\` to check your account's usage and remaining quota.`
+          }
+        }
+      ]
+    })
+  );
+}
+
+// src/prompts/write-policy.ts
+var import_zod5 = require("zod");
+function registerWritePolicy(server) {
+  server.prompt(
+    "write-policy",
+    "Guide for writing a Vettly moderation policy in YAML. Optionally tailored to a specific platform type.",
+    { platform: import_zod5.z.string().optional().describe('The type of platform (e.g. "kids app", "marketplace", "social media", "dating app")') },
+    async (args) => {
+      const platformContext = args.platform ? `
+
+The user is building a **${args.platform}**. Tailor your policy suggestions to this platform type \u2014 consider what content categories matter most and what thresholds are appropriate.` : "";
+      return {
+        messages: [
+          {
+            role: "user",
+            content: {
+              type: "text",
+              text: `You are helping a user write a Vettly moderation policy in YAML.${platformContext}
+
+## Policy YAML structure
+
+A Vettly policy is a YAML document that defines moderation rules in plain English. Here's the structure:
+
+\`\`\`yaml
+name: My Policy
+description: Brief description of what this policy covers
+
+rules:
+  - category: hate_speech
+    threshold: 0.7
+    action: block
+    description: Block hateful content targeting protected groups
+
+  - category: violence
+    threshold: 0.8
+    action: flag
+    description: Flag violent content for human review
+
+  - category: sexual_content
+    threshold: 0.5
+    action: block
+    description: Block explicit sexual content
+
+  - category: self_harm
+    threshold: 0.6
+    action: block
+    description: Block content promoting self-harm
+\`\`\`
+
+## Available categories
+Common categories: \`hate_speech\`, \`violence\`, \`sexual_content\`, \`self_harm\`, \`harassment\`, \`dangerous_content\`, \`spam\`
+
+## Actions
+- \`allow\` \u2014 let the content through
+- \`flag\` \u2014 mark for human review
+- \`block\` \u2014 reject the content
+
+## Thresholds
+- 0.0\u20130.3: Very sensitive (catches more, higher false positive rate)
+- 0.4\u20130.6: Moderate sensitivity
+- 0.7\u20131.0: High confidence only (fewer false positives)
+
+## Workflow
+1. Draft the YAML policy based on the user's requirements
+2. Use the \`validate_policy\` tool to check for errors
+3. Iterate until validation passes
+4. The user can then save the policy via the Vettly dashboard`
+            }
+          }
+        ]
+      };
+    }
+  );
+}
+
+// src/prompts/review-flagged-content.ts
+var import_zod6 = require("zod");
+function registerReviewFlaggedContent(server) {
+  server.prompt(
+    "review-flagged-content",
+    "Template for reviewing flagged moderation decisions. Optionally filtered to a specific policy.",
+    { policyId: import_zod6.z.string().optional().describe("Filter flagged decisions to a specific policy ID") },
+    async (args) => {
+      const policyFilter = args.policyId ? `
+
+Filter decisions to policy ID: \`${args.policyId}\`` : "";
+      return {
+        messages: [
+          {
+            role: "user",
+            content: {
+              type: "text",
+              text: `You are helping a user review flagged moderation decisions in Vettly.${policyFilter}
+
+## Review workflow
+
+### Step 1: Fetch flagged decisions
+Use the \`get_recent_decisions\` tool with \`flagged: true\` to get recent flagged content.${args.policyId ? ` Also set \`policyId: "${args.policyId}"\` to filter by that policy.` : " Optionally filter by `policyId` or `contentType`."}
+
+### Step 2: Analyze each decision
+For each flagged decision, examine:
+- **Content**: What was submitted
+- **Categories**: Which categories triggered and their scores
+- **Action**: What action was taken (flag vs block)
+- **Policy**: Which policy was applied
+
+### Step 3: Provide recommendations
+For each flagged item, suggest one of:
+- **Approve**: The content is acceptable, the flag was a false positive
+- **Confirm block**: The content correctly violates the policy
+- **Adjust policy**: The threshold may need tuning (too sensitive or not enough)
+
+### Step 4: Check for patterns
+Use \`get_usage_stats\` to look at overall moderation trends. If many items are being flagged, the policy thresholds may need adjustment.
+
+## Tips
+- High category scores (>0.9) are almost always correct flags
+- Scores in the 0.5-0.7 range are where most false positives occur
+- If a category consistently flags incorrectly, raise its threshold in the policy`
+            }
+          }
+        ]
+      };
+    }
+  );
+}
+
+// src/prompts/index.ts
+function registerPrompts(server) {
+  registerModerateContentGuide(server);
+  registerWritePolicy(server);
+  registerReviewFlaggedContent(server);
+}
+
 // src/server.ts
 function createVettlyMcpServer(config) {
   const client = new import_sdk.ModerationClient({
@@ -328,6 +539,7 @@ function createVettlyMcpServer(config) {
   });
   registerTools(server, client);
   registerResources(server, client);
+  registerPrompts(server);
   return server;
 }
 

package/dist/bin/vettly-mcp.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   createVettlyMcpServer
-} from "../chunk-VZJIPBOX.js";
+} from "../chunk-L5ZHIGXM.js";
 
 // src/bin/vettly-mcp.ts
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";

package/dist/{chunk-VZJIPBOX.js → chunk-L5ZHIGXM.js}

@@ -18,6 +18,13 @@ function registerModerateTool(server, client) {
       policyId: ModerateInputSchema.shape.policyId,
       contentType: ModerateInputSchema.shape.contentType
     },
+    {
+      title: "Moderate Content",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = ModerateInputSchema.parse(args);
       try {
@@ -74,6 +81,13 @@ function registerValidateTool(server, client) {
     {
       yamlContent: ValidateInputSchema.shape.yamlContent
     },
+    {
+      title: "Validate Policy",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = ValidateInputSchema.parse(args);
       try {
@@ -107,6 +121,13 @@ function registerListPoliciesTool(server, client) {
     "list_policies",
     "List all moderation policies available in your Vettly account.",
     {},
+    {
+      title: "List Policies",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async () => {
       try {
         const result = await client.listPolicies();
@@ -145,6 +166,13 @@ function registerUsageStatsTool(server, client) {
     {
       days: UsageStatsInputSchema.shape.days
     },
+    {
+      title: "Get Usage Stats",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = UsageStatsInputSchema.parse(args);
       try {
@@ -190,6 +218,13 @@ function registerDecisionsTool(server, client) {
       policyId: DecisionsInputSchema.shape.policyId,
       contentType: DecisionsInputSchema.shape.contentType
     },
+    {
+      title: "Get Recent Decisions",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = DecisionsInputSchema.parse(args);
       try {
@@ -310,6 +345,182 @@ function registerResources(server, client) {
   registerPolicyResources(server, client);
 }
 
+// src/prompts/moderate-content-guide.ts
+function registerModerateContentGuide(server) {
+  server.prompt(
+    "moderate-content-guide",
+    "Step-by-step guide to moderating content with Vettly. Explains the workflow: list policies, moderate content, and interpret results.",
+    async () => ({
+      messages: [
+        {
+          role: "user",
+          content: {
+            type: "text",
+            text: `You are helping a user moderate content with Vettly. Follow this workflow:
+
+## Step 1: List available policies
+Use the \`list_policies\` tool to see what moderation policies are configured. Each policy has an ID and defines rules for what content is allowed or flagged.
+
+## Step 2: Moderate content
+Use the \`moderate_content\` tool with:
+- \`content\`: The text, image URL, or video URL to check
+- \`policyId\`: The policy ID from step 1
+- \`contentType\`: "text", "image", or "video"
+
+## Step 3: Interpret results
+The response includes:
+- \`safe\`: Boolean \u2014 whether the content passed moderation
+- \`flagged\`: Boolean \u2014 whether the content was flagged for review
+- \`action\`: The recommended action ("allow", "flag", or "block")
+- \`categories\`: Breakdown of category scores (e.g. hate speech, violence, sexual content)
+- \`decisionId\`: Unique ID for this decision, useful for auditing
+
+## Tips
+- Always list policies first so you know which policy ID to use.
+- For images and video, pass the URL as the \`content\` parameter.
+- Use \`get_recent_decisions\` to review past moderation outcomes.
+- Use \`get_usage_stats\` to check your account's usage and remaining quota.`
+          }
+        }
+      ]
+    })
+  );
+}
+
+// src/prompts/write-policy.ts
+import { z as z5 } from "zod";
+function registerWritePolicy(server) {
+  server.prompt(
+    "write-policy",
+    "Guide for writing a Vettly moderation policy in YAML. Optionally tailored to a specific platform type.",
+    { platform: z5.string().optional().describe('The type of platform (e.g. "kids app", "marketplace", "social media", "dating app")') },
+    async (args) => {
+      const platformContext = args.platform ? `
+
+The user is building a **${args.platform}**. Tailor your policy suggestions to this platform type \u2014 consider what content categories matter most and what thresholds are appropriate.` : "";
+      return {
+        messages: [
+          {
+            role: "user",
+            content: {
+              type: "text",
+              text: `You are helping a user write a Vettly moderation policy in YAML.${platformContext}
+
+## Policy YAML structure
+
+A Vettly policy is a YAML document that defines moderation rules in plain English. Here's the structure:
+
+\`\`\`yaml
+name: My Policy
+description: Brief description of what this policy covers
+
+rules:
+  - category: hate_speech
+    threshold: 0.7
+    action: block
+    description: Block hateful content targeting protected groups
+
+  - category: violence
+    threshold: 0.8
+    action: flag
+    description: Flag violent content for human review
+
+  - category: sexual_content
+    threshold: 0.5
+    action: block
+    description: Block explicit sexual content
+
+  - category: self_harm
+    threshold: 0.6
+    action: block
+    description: Block content promoting self-harm
+\`\`\`
+
+## Available categories
+Common categories: \`hate_speech\`, \`violence\`, \`sexual_content\`, \`self_harm\`, \`harassment\`, \`dangerous_content\`, \`spam\`
+
+## Actions
+- \`allow\` \u2014 let the content through
+- \`flag\` \u2014 mark for human review
+- \`block\` \u2014 reject the content
+
+## Thresholds
+- 0.0\u20130.3: Very sensitive (catches more, higher false positive rate)
+- 0.4\u20130.6: Moderate sensitivity
+- 0.7\u20131.0: High confidence only (fewer false positives)
+
+## Workflow
+1. Draft the YAML policy based on the user's requirements
+2. Use the \`validate_policy\` tool to check for errors
+3. Iterate until validation passes
+4. The user can then save the policy via the Vettly dashboard`
+            }
+          }
+        ]
+      };
+    }
+  );
+}
+
+// src/prompts/review-flagged-content.ts
+import { z as z6 } from "zod";
+function registerReviewFlaggedContent(server) {
+  server.prompt(
+    "review-flagged-content",
+    "Template for reviewing flagged moderation decisions. Optionally filtered to a specific policy.",
+    { policyId: z6.string().optional().describe("Filter flagged decisions to a specific policy ID") },
+    async (args) => {
+      const policyFilter = args.policyId ? `
+
+Filter decisions to policy ID: \`${args.policyId}\`` : "";
+      return {
+        messages: [
+          {
+            role: "user",
+            content: {
+              type: "text",
+              text: `You are helping a user review flagged moderation decisions in Vettly.${policyFilter}
+
+## Review workflow
+
+### Step 1: Fetch flagged decisions
+Use the \`get_recent_decisions\` tool with \`flagged: true\` to get recent flagged content.${args.policyId ? ` Also set \`policyId: "${args.policyId}"\` to filter by that policy.` : " Optionally filter by `policyId` or `contentType`."}
+
+### Step 2: Analyze each decision
+For each flagged decision, examine:
+- **Content**: What was submitted
+- **Categories**: Which categories triggered and their scores
+- **Action**: What action was taken (flag vs block)
+- **Policy**: Which policy was applied
+
+### Step 3: Provide recommendations
+For each flagged item, suggest one of:
+- **Approve**: The content is acceptable, the flag was a false positive
+- **Confirm block**: The content correctly violates the policy
+- **Adjust policy**: The threshold may need tuning (too sensitive or not enough)
+
+### Step 4: Check for patterns
+Use \`get_usage_stats\` to look at overall moderation trends. If many items are being flagged, the policy thresholds may need adjustment.
+
+## Tips
+- High category scores (>0.9) are almost always correct flags
+- Scores in the 0.5-0.7 range are where most false positives occur
+- If a category consistently flags incorrectly, raise its threshold in the policy`
+            }
+          }
+        ]
+      };
+    }
+  );
+}
+
+// src/prompts/index.ts
+function registerPrompts(server) {
+  registerModerateContentGuide(server);
+  registerWritePolicy(server);
+  registerReviewFlaggedContent(server);
+}
+
 // src/server.ts
 function createVettlyMcpServer(config) {
   const client = new ModerationClient({
@@ -322,6 +533,7 @@ function createVettlyMcpServer(config) {
   });
   registerTools(server, client);
   registerResources(server, client);
+  registerPrompts(server);
   return server;
 }
 

package/dist/index.cjs

@@ -20,9 +20,11 @@ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: tru
 // src/index.ts
 var index_exports = {};
 __export(index_exports, {
+  configSchema: () => configSchema,
   createVettlyMcpServer: () => createVettlyMcpServer
 });
 module.exports = __toCommonJS(index_exports);
+var import_zod7 = require("zod");
 
 // src/server.ts
 var import_mcp2 = require("@modelcontextprotocol/sdk/server/mcp.js");
@@ -44,6 +46,13 @@ function registerModerateTool(server, client) {
       policyId: ModerateInputSchema.shape.policyId,
       contentType: ModerateInputSchema.shape.contentType
     },
+    {
+      title: "Moderate Content",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = ModerateInputSchema.parse(args);
       try {
@@ -100,6 +109,13 @@ function registerValidateTool(server, client) {
     {
       yamlContent: ValidateInputSchema.shape.yamlContent
     },
+    {
+      title: "Validate Policy",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = ValidateInputSchema.parse(args);
       try {
@@ -133,6 +149,13 @@ function registerListPoliciesTool(server, client) {
     "list_policies",
     "List all moderation policies available in your Vettly account.",
     {},
+    {
+      title: "List Policies",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async () => {
       try {
         const result = await client.listPolicies();
@@ -171,6 +194,13 @@ function registerUsageStatsTool(server, client) {
     {
       days: UsageStatsInputSchema.shape.days
     },
+    {
+      title: "Get Usage Stats",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = UsageStatsInputSchema.parse(args);
       try {
@@ -216,6 +246,13 @@ function registerDecisionsTool(server, client) {
       policyId: DecisionsInputSchema.shape.policyId,
       contentType: DecisionsInputSchema.shape.contentType
     },
+    {
+      title: "Get Recent Decisions",
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
     async (args) => {
       const input = DecisionsInputSchema.parse(args);
       try {
@@ -336,6 +373,182 @@ function registerResources(server, client) {
   registerPolicyResources(server, client);
 }
 
+// src/prompts/moderate-content-guide.ts
+function registerModerateContentGuide(server) {
+  server.prompt(
+    "moderate-content-guide",
+    "Step-by-step guide to moderating content with Vettly. Explains the workflow: list policies, moderate content, and interpret results.",
+    async () => ({
+      messages: [
+        {
+          role: "user",
+          content: {
+            type: "text",
+            text: `You are helping a user moderate content with Vettly. Follow this workflow:
+
+## Step 1: List available policies
+Use the \`list_policies\` tool to see what moderation policies are configured. Each policy has an ID and defines rules for what content is allowed or flagged.
+
+## Step 2: Moderate content
+Use the \`moderate_content\` tool with:
+- \`content\`: The text, image URL, or video URL to check
+- \`policyId\`: The policy ID from step 1
+- \`contentType\`: "text", "image", or "video"
+
+## Step 3: Interpret results
+The response includes:
+- \`safe\`: Boolean \u2014 whether the content passed moderation
+- \`flagged\`: Boolean \u2014 whether the content was flagged for review
+- \`action\`: The recommended action ("allow", "flag", or "block")
+- \`categories\`: Breakdown of category scores (e.g. hate speech, violence, sexual content)
+- \`decisionId\`: Unique ID for this decision, useful for auditing
+
+## Tips
+- Always list policies first so you know which policy ID to use.
+- For images and video, pass the URL as the \`content\` parameter.
+- Use \`get_recent_decisions\` to review past moderation outcomes.
+- Use \`get_usage_stats\` to check your account's usage and remaining quota.`
+          }
+        }
+      ]
+    })
+  );
+}
+
+// src/prompts/write-policy.ts
+var import_zod5 = require("zod");
+function registerWritePolicy(server) {
+  server.prompt(
+    "write-policy",
+    "Guide for writing a Vettly moderation policy in YAML. Optionally tailored to a specific platform type.",
+    { platform: import_zod5.z.string().optional().describe('The type of platform (e.g. "kids app", "marketplace", "social media", "dating app")') },
+    async (args) => {
+      const platformContext = args.platform ? `
+
+The user is building a **${args.platform}**. Tailor your policy suggestions to this platform type \u2014 consider what content categories matter most and what thresholds are appropriate.` : "";
+      return {
+        messages: [
+          {
+            role: "user",
+            content: {
+              type: "text",
+              text: `You are helping a user write a Vettly moderation policy in YAML.${platformContext}
+
+## Policy YAML structure
+
+A Vettly policy is a YAML document that defines moderation rules in plain English. Here's the structure:
+
+\`\`\`yaml
+name: My Policy
+description: Brief description of what this policy covers
+
+rules:
+  - category: hate_speech
+    threshold: 0.7
+    action: block
+    description: Block hateful content targeting protected groups
+
+  - category: violence
+    threshold: 0.8
+    action: flag
+    description: Flag violent content for human review
+
+  - category: sexual_content
+    threshold: 0.5
+    action: block
+    description: Block explicit sexual content
+
+  - category: self_harm
+    threshold: 0.6
+    action: block
+    description: Block content promoting self-harm
+\`\`\`
+
+## Available categories
+Common categories: \`hate_speech\`, \`violence\`, \`sexual_content\`, \`self_harm\`, \`harassment\`, \`dangerous_content\`, \`spam\`
+
+## Actions
+- \`allow\` \u2014 let the content through
+- \`flag\` \u2014 mark for human review
+- \`block\` \u2014 reject the content
+
+## Thresholds
+- 0.0\u20130.3: Very sensitive (catches more, higher false positive rate)
+- 0.4\u20130.6: Moderate sensitivity
+- 0.7\u20131.0: High confidence only (fewer false positives)
+
+## Workflow
+1. Draft the YAML policy based on the user's requirements
+2. Use the \`validate_policy\` tool to check for errors
+3. Iterate until validation passes
+4. The user can then save the policy via the Vettly dashboard`
+            }
+          }
+        ]
+      };
+    }
+  );
+}
+
+// src/prompts/review-flagged-content.ts
+var import_zod6 = require("zod");
+function registerReviewFlaggedContent(server) {
+  server.prompt(
+    "review-flagged-content",
+    "Template for reviewing flagged moderation decisions. Optionally filtered to a specific policy.",
+    { policyId: import_zod6.z.string().optional().describe("Filter flagged decisions to a specific policy ID") },
+    async (args) => {
+      const policyFilter = args.policyId ? `
+
+Filter decisions to policy ID: \`${args.policyId}\`` : "";
+      return {
+        messages: [
+          {
+            role: "user",
+            content: {
+              type: "text",
+              text: `You are helping a user review flagged moderation decisions in Vettly.${policyFilter}
+
+## Review workflow
+
+### Step 1: Fetch flagged decisions
+Use the \`get_recent_decisions\` tool with \`flagged: true\` to get recent flagged content.${args.policyId ? ` Also set \`policyId: "${args.policyId}"\` to filter by that policy.` : " Optionally filter by `policyId` or `contentType`."}
+
+### Step 2: Analyze each decision
+For each flagged decision, examine:
+- **Content**: What was submitted
+- **Categories**: Which categories triggered and their scores
+- **Action**: What action was taken (flag vs block)
+- **Policy**: Which policy was applied
+
+### Step 3: Provide recommendations
+For each flagged item, suggest one of:
+- **Approve**: The content is acceptable, the flag was a false positive
+- **Confirm block**: The content correctly violates the policy
+- **Adjust policy**: The threshold may need tuning (too sensitive or not enough)
+
+### Step 4: Check for patterns
+Use \`get_usage_stats\` to look at overall moderation trends. If many items are being flagged, the policy thresholds may need adjustment.
+
+## Tips
+- High category scores (>0.9) are almost always correct flags
+- Scores in the 0.5-0.7 range are where most false positives occur
+- If a category consistently flags incorrectly, raise its threshold in the policy`
+            }
+          }
+        ]
+      };
+    }
+  );
+}
+
+// src/prompts/index.ts
+function registerPrompts(server) {
+  registerModerateContentGuide(server);
+  registerWritePolicy(server);
+  registerReviewFlaggedContent(server);
+}
+
 // src/server.ts
 function createVettlyMcpServer(config) {
   const client = new import_sdk.ModerationClient({
@@ -348,9 +561,19 @@ function createVettlyMcpServer(config) {
   });
   registerTools(server, client);
   registerResources(server, client);
+  registerPrompts(server);
   return server;
 }
+
+// src/index.ts
+var configSchema = import_zod7.z.object({
+  vettlyApiKey: import_zod7.z.string().describe(
+    "Your Vettly API key (starts with vettly_live_ or vettly_test_). Get one free at https://vettly.dev"
+  ),
+  apiUrl: import_zod7.z.string().url().default("https://api.vettly.dev").describe("Custom API base URL. Only change this for self-hosted Vettly instances.")
+});
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
+  configSchema,
   createVettlyMcpServer
 });

package/dist/index.js

@@ -1,6 +1,16 @@
 import {
   createVettlyMcpServer
-} from "./chunk-VZJIPBOX.js";
+} from "./chunk-L5ZHIGXM.js";
+
+// src/index.ts
+import { z } from "zod";
+var configSchema = z.object({
+  vettlyApiKey: z.string().describe(
+    "Your Vettly API key (starts with vettly_live_ or vettly_test_). Get one free at https://vettly.dev"
+  ),
+  apiUrl: z.string().url().default("https://api.vettly.dev").describe("Custom API base URL. Only change this for self-hosted Vettly instances.")
+});
 export {
+  configSchema,
   createVettlyMcpServer
 };

package/package.json

@@ -1,8 +1,8 @@
 {
   "name": "@vettly/mcp",
-  "version": "0.1.7",
+  "version": "0.1.9",
   "mcpName": "io.github.code-with-brian/vettly",
-  "description": "MCP server for content moderation with Claude and Cursor. App Store policy testing.",
+  "description": "MCP server for content moderation with Claude and Cursor. Content moderation policy testing.",
   "type": "module",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",
@@ -18,7 +18,8 @@
   },
   "files": [
     "dist",
-    "README.md"
+    "README.md",
+    "smithery.yaml"
   ],
   "scripts": {
     "build": "tsup src/index.ts src/bin/vettly-mcp.ts --format esm,cjs --dts --clean",

package/smithery.yaml

@@ -0,0 +1,21 @@
+# Smithery configuration file: https://smithery.ai/docs/config#smitheryyaml
+
+startCommand:
+  type: stdio
+
+configSchema:
+  type: object
+  required:
+    - vettlyApiKey
+  properties:
+    vettlyApiKey:
+      type: string
+      description: Your Vettly API key (starts with vettly_live_ or vettly_test_). Get one free at https://vettly.dev
+    apiUrl:
+      type: string
+      default: https://api.vettly.dev
+      description: Custom API base URL. Only change this for self-hosted Vettly instances.
+
+commandFunction:
+  |-
+  (config) => ({ command: 'npx', args: ['-y', '@vettly/mcp'], env: { VETTLY_API_KEY: config.vettlyApiKey, ...(config.apiUrl && config.apiUrl !== 'https://api.vettly.dev' ? { VETTLY_API_URL: config.apiUrl } : {}) } })

package/dist/bin/vettly-mcp.d.cts

@@ -1 +0,0 @@
-#!/usr/bin/env node

package/dist/bin/vettly-mcp.d.ts

@@ -1 +0,0 @@
-#!/usr/bin/env node

package/dist/index.d.cts

@@ -1,9 +0,0 @@
-import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
-
-interface VettlyMcpConfig {
-    apiKey: string;
-    apiUrl?: string;
-}
-declare function createVettlyMcpServer(config: VettlyMcpConfig): McpServer;
-
-export { type VettlyMcpConfig, createVettlyMcpServer };

package/dist/index.d.ts

@@ -1,9 +0,0 @@
-import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
-
-interface VettlyMcpConfig {
-    apiKey: string;
-    apiUrl?: string;
-}
-declare function createVettlyMcpServer(config: VettlyMcpConfig): McpServer;
-
-export { type VettlyMcpConfig, createVettlyMcpServer };