Feature/drel 817 migrate mcp from ipfc cids to npm pkg names

.nx/version-plans/version-plan-1751658342657.md

@@ -0,0 +1,7 @@
+---
+app-sdk: minor
+mcp-sdk: major
+mcp: major
+---
+
+Migration of MCP building from JSON file exclusively to Registry based with json available for overrides

docs/src/Developers/App-Agent-Developers/MCP.md

@@ -7,9 +7,9 @@ title: MCP - Model Context Protocol
 
 Any Vincent App can be converted into a Model Protocol Server (MCP) that can be consumed by any Large Language Model (LLM) with support for the MCP standard.
 
-We provide an [implementation of an MCP Server](https://github.com/LIT-Protocol/Vincent/tree/feature/main/packages/apps/mcp), connectable through STDIO and HTTP transports. You can use it (forking or [using `npx`](https://www.npmjs.com/package/@lit-protocol/vincent-mcp-server)) with your keys and Vincent Apps or you can customize the whole process to make your own Vincent MCP Server.
+We provide an [implementation of an MCP Server](https://github.com/LIT-Protocol/Vincent/tree/feature/main/packages/apps/mcp), connectable through STDIO and HTTP transports. You can use it (forking or [using `npx`](https://www.npmjs.com/package/@lit-protocol/vincent-mcp-server)) with your keys and Vincent Apps or you can customize the whole process to make your own Vincent MCP Server based on our [MCP SDK](https://github.com/LIT-Protocol/Vincent/blob/main/packages/libs/mcp-sdk/README.md).
 
-By following this process, your Vincent App tools will be exposed to LLMs as a set of MCP tools. The MCP server can also be extended with custom tools and prompts to suit your specific needs.
+By following this process, your Vincent App tools will be exposed to LLMs as a set of MCP tools. The MCP server can also be extended with custom tools and prompts to suit your specific needs or provide extra capabilities around the Vincent Tool ecosystem.
 
 And if you're building an AI application, check out our [OpenAI AgentKit demo](https://github.com/LIT-Protocol/Vincent-MCP-OpenAI-AgentKit) for a guide on how to integrate your Vincent App with the OpenAI AgentKit.
 
@@ -34,7 +34,8 @@ const appDef: VincentAppDef = {
   name: 'My Vincent App',
   description: 'A Vincent application that executes tools for its delegators',
   tools: {
-    QmIpfsCid1: {
+    '@organization/npm-published-vincent-tool': {
+      version: '1.0.0',
       name: 'myTool',
       description: 'A tool that does something',
       parameters: [
@@ -55,8 +56,6 @@ const appDef: VincentAppDef = {
 const server = await getVincentAppServer(wallet, appDef);
 ```
 
-You can check the [Uniswap Swap example app json](https://github.com/LIT-Protocol/Vincent/blob/feature/main/packages/apps/mcp/vincent-app.example.json) for a complete Vincent App definition.
-
 ## Extending the MCP Server
 
 At this moment you can add more tools, resources or prompts to the server.
@@ -67,7 +66,7 @@ server.resource(...);
 server.prompt(...);
 ```
 
-These tools, resources and prompts will be exposed in the server along with the ones from the Vincent App definition. Consider adding any other tools that you want to be executed by the LLM and that are not Vincent Tools such as tools to query balance or fetch useful data from external sources.
+These tools, resources and prompts will be exposed in the server along with the ones from the Vincent App definition. Consider adding any other tools that you want to be executed by the LLM and that are not Vincent Tools. For example, you could add tools to query balance or fetch useful data from external sources.
 
 ## Picking a Transport
 
@@ -126,7 +125,7 @@ This MCP Server includes:
 
 - HTTP and STDIO transports
 - `.env` file support for environment definition
-- App definition with a custom JSON file to define which tools and params are exposed to LLMs
+- App definition overriding with a custom JSON file to refine descriptions and filter tools to be exposed to LLMs
 - Support for delegatee and delegators in HTTP transport
   - Delegators MUST authenticate with SIWE OR their Vincent JWT
   - Delegatees MUST identify with SIWE
@@ -162,7 +161,7 @@ Before deploying, you'll need to create the following two files in the root of y
     CMD ["npx", "@lit-protocol/vincent-mcp-server", "http"]
     ```
 
-2.  Create the Vincent App JSON definition file. Fill it with the data of your Vincent: ID, version, name, description and tools data. Check the [Uniswap Swap example app json](https://github.com/LIT-Protocol/Vincent/blob/feature/main/packages/apps/mcp/vincent-app.example.json) for a complete Vincent App definition.
+2.  Create the Vincent App JSON definition override file. Fill it with the data of your Vincent App you want to override: ID, version, name, description and tools data. Check the [Uniswap Swap example app json](https://github.com/LIT-Protocol/Vincent/blob/feature/main/packages/apps/mcp/vincent-app.example.json) for a complete Vincent App override file.
 
 3.  Add both files to git. Commit and push them to your repository to use as source for Heroku or Render.
 

packages/apps/mcp/README.md

@@ -6,49 +6,45 @@ It leverages the `@lit-protocol/vincent-mcp-sdk` to build a server from a Vincen
 
 ## Setup
 
-- Copy `vincent-app.example.json` to `vincent-app.json` or any other name you want and configure your Vincent App definition in it.
-- Copy `.env.example` to `.env` and fill in the values. Use absolute paths for the `VINCENT_APP_JSON_DEFINITION` value.
+- Optional: Copy `vincent-app.example.json` to `vincent-app.json` or any other name you want and configure your Vincent App definition overrides in it. If no overrides are needed, then this file can be omitted.
+- Optional: Copy `.env.example` to `.env` and fill in the values. Use absolute paths for the `VINCENT_APP_JSON_DEFINITION` value. You can also pass them via CLI arguments when calling the script.
 
-# Writing App definition JSON file
+# Writing App definition overrides in the JSON file
 
-To define the Vincent App that will be transformed into an MCP Server, a JSON definition of it must be provided.
+Name and descriptions provided by developers in the registry might not be very descriptive to LLMs or you may want to modify them.
+
+In order to override any of those values, create a `.json` file with the following structure:
 
 ```json
 {
   "id": "8462368", // The Id of the Vincent App
   "version": "1", // The version of the Vincent App
   "name": "My Vincent App", // Name of the Vincent App. Can be overriden, doesn't have to be the same as in the registry.
   "description": "A Vincent application that executes tools for its delegators", // Description of the Vincent App. Can be overriden, doesn't have to be the same as in the registry.
+  // Adding a tools object will override what is already present in the registry. Without this field, all tools from the registry will be exposed.
   "tools": {
-    // Any tool that you want to expose to the LLM has to be included using its IPFS CID as key in this object. If a tool is not included, it is not exposed as an MCP Server tool.
-    "QmIpfsCid1": {
-      "name": "myTool", // Name of the tool. Can be overriden, doesn't have to be the same as in the registry.
-      "description": "A tool that does something", // Description of the tool. Can be overriden, doesn't have to be the same as in the registry.
-      // All parameters of the tool have to be added under this array or the LLM won't be able to see them or provide values for it
+    // Any tool that you want to expose to the LLM has to be included using its NPM package name as key in this object. If a tool is not included, it is not exposed as an MCP Server tool.
+    "vincent-tool-npm-pkg-name": {
+      "name": "myTool", // Name of the tool. Defaults to npm pkg name.
+      "description": "A tool that does something", // Description of the tool.
       "parameters": [
         {
-          "name": "param1", // Name of the param. Cannot be overriden.
-          "type": "string", // Type of the param. Must be the type the tool expects.
-          "description": "A parameter that is used in the tool to do something" // Description of the param. Can be overriden.
+          "name": "param1", // Name of the param. Used to identify and apply the rest of properties.
+          "description": "A parameter that is used in the tool to do something" // Description of the param.
         }
-        // ...rest of params you want to expose.
-        // Any optional param that is not included here will be exposed by the tool.
+        // ...rest of params you want to override.
       ]
-    }
+    },
+    "vincent-tool-without-overrides": {} // Empty objects mean that the tool is exposed but with default values.
   }
 }
 ```
 
-For any value that can be overriden, consider that those are the hints the LLM uses to know how to use the tool. Therefore, those are good places to provide any information you want the LLM to know about the tool such as units, formats, examples or pre-conditions to check.
-
-If you are the owner of the app, most of the data can be obtained from the Vincent App page in the Vincent dashboard.
+When the `tools` property is omitted, all tools from the registry will be exposed. So when overriding at least one tool, you need to specify all others that you want to still expose as MCP tools, even with empty values.
 
-If you are not the owner of the app, the tool fields and its included tools IPFS CIDs are shown in the consent screen.
-
-The IPFS CID can also be obtained from the bundled tool code published in npm. For example [vincent-tool-metadata.json](../tool-erc20-approval/src/generated/vincent-tool-metadata.json) for our ERC20 approval tool.
-To get the tool params from source code, you can check the tool schemas such as [schemas.ts](../tool-erc20-approval/src/lib/schemas.ts) for our ERC20 approval tool.
+For any value that can be overriden, consider that those are the hints the LLM uses to know how to use the tool. Therefore, those are good places to provide any information you want the LLM to know about the tool such as units, formats, examples or pre-conditions to check.
 
-Any tool created using our [Tools and Policies SDK](https://www.npmjs.com/package/@lit-protocol/vincent-tool-sdk) will provide those files.
+Tools included in definition MUST be published in NPM and imported using their package names. Also they must be part of the Vincent App and already registered. Any tool that is not part of that specific app will fail its invocation.
 
 # Running
 
@@ -66,10 +62,13 @@ You can run the Vincent MCP server directly using npx without downloading the re
 npx @lit-protocol/vincent-mcp-server stdio
 ```
 
-When setting this in the LLM client, pass it the necessary environment variables from your client. These env variables include:
+When setting this in the LLM client, pass it the necessary environment variables from your LLM client. These env variables include:
 
-- `VINCENT_APP_JSON_DEFINITION`: Path to your Vincent App definition JSON file
-- `VINCENT_DELEGATEE_PRIVATE_KEY`: The private key of the delegatee. This is the one you added in the Vincent App Dashboard as [an authorized signer for your app](https://docs.heyvincent.ai/documents/Quick_Start.html#:~:text=New%20App%22%20button.-,Delegatees,-%3A%20Delegatees%20are). This private key MUST be an allowed delegatee of the Vincent App defined in the JSON.
+- `VINCENT_DELEGATEE_PRIVATE_KEY`: The private key of the delegatee. This is the one you added in the Vincent App Dashboard as [an authorized signer for your app](https://docs.heyvincent.ai/documents/Quick_Start.html#:~:text=New%20App%22%20button.-,Delegatees,-%3A%20Delegatees%20are). This private key MUST be an allowed delegatee of the Vincent App.
+- (Optional) `VINCENT_APP_ID`: The Vincent App Id you want to run as an MCP Server
+- (Optional) `VINCENT_APP_JSON_DEFINITION`: Path to your Vincent App overrides JSON file
+
+Either in `VINCENT_APP_ID` or in the `VINCENT_APP_JSON_DEFINITION` file, the App Id MUST be specified. So one of those values is required.
 
 ### HTTP mode
 
@@ -79,28 +78,29 @@ npx @lit-protocol/vincent-mcp-server http
 
 In HTTP mode, the environment variables are configured on the server itself, not the client running it.
 
-These commands require the following environment variables to be set:
+To configure runtime environment in this mode, set the following environment variables:
 
+- `VINCENT_DELEGATEE_PRIVATE_KEY`: The private key of the delegatee. This is the one you added in the Vincent App Dashboard as [an authorized signer for your app](https://docs.heyvincent.ai/documents/Quick_Start.html#:~:text=New%20App%22%20button.-,Delegatees,-%3A%20Delegatees%20are). This private key MUST be an allowed delegatee of the Vincent App.
+- (Optional) `VINCENT_APP_ID`: The Vincent App Id you want to run as an MCP Server
+- (Optional) `VINCENT_APP_JSON_DEFINITION`: Path to your Vincent App overrides JSON file
 - `EXPECTED_AUDIENCE`: The audience that you expect JWTs to have. Vincent populates this with the redirect URLs. Likely you want this server to be one of those URLs.
-- `VINCENT_APP_JSON_DEFINITION`: Path to your Vincent App definition JSON file
-- `VINCENT_DELEGATEE_PRIVATE_KEY`: The private key of the delegatee. This is the one you added in the Vincent App Dashboard as [an authorized signer for your app](https://docs.heyvincent.ai/documents/Quick_Start.html#:~:text=New%20App%22%20button.-,Delegatees,-%3A%20Delegatees%20are).
-- `VINCENT_MCP_BASE_URL`: This MCP server URL
-- `PORT` (for HTTP mode only): The port to run the HTTP server on (defaults to 3000)
-
-Other optional environment variables include:
+- `VINCENT_MCP_BASE_URL`: This MCP server URL. Used to generate SIWE messages and verify signatures
+- `VINCENT_REGISTRY_URL`: This Vincent Registry server URL. Will be queried to get the Vincent App and its tools info
+- (Optional) `PORT`: The port to run the HTTP server on (defaults to 3000)
+- (Optional) `HTTP_TRANSPORT_CLEAN_INTERVAL`: Defines the interval (milliseconds) that the server will use to clean unused transports. Defaults to 1 hour
+- (Optional) `HTTP_TRANSPORT_TTL`: Defines the time (milliseconds) that a transport will still be considered in use after the last time it was actually used. Defaults to 1 hour
+- (Optional) `SIWE_EXPIRATION_TIME`: Duration of the generated SIWE message to sign. Defaults to 1 hour
+- (Optional) `SIWE_NONCE_CLEAN_INTERVAL`: Defines the interval (milliseconds) that the server will use to clean unused transports. Defaults to 1 hour
+- (Optional) `SIWE_NONCE_TTL`: Defines the time (milliseconds) that a SIWE nonce will still be considered valid after it was created. Defaults to 5 minutes
 
-- `HTTP_TRANSPORT_CLEAN_INTERVAL`: Defines the interval (milliseconds) that the server will use to clean unused transports. Defaults to 1 hour
-- `HTTP_TRANSPORT_TTL`: Defines the time (milliseconds) that a transport will still be considered in use after the last time it was actually used. Defaults to 1 hour
-- `SIWE_EXPIRATION_TIME`: Duration of the generated SIWE message to sign. Defaults to 1 hour
-- `SIWE_NONCE_CLEAN_INTERVAL`: Defines the interval (milliseconds) that the server will use to clean unused transports. Defaults to 1 hour
-- `SIWE_NONCE_TTL`: Defines the time (milliseconds) that a SIWE nonce will still be considered valid after it was created. Defaults to 5 minutes
+Either in `VINCENT_APP_ID` or in the `VINCENT_APP_JSON_DEFINITION` file, the App Id MUST be specified. So one of those values is required.
 
 Consider that a SIWE message must have a valid nonce, so it will become invalid after reaching the expiration time or the nonce has been discarded.
 
 You can set these environment variables in your shell before running the commands, or use a tool like `dotenvx`:
 
 ```bash
-dotenvx run -f /path/to/.env -- npx @lit-protocol/vincent-mcp-server http
+dotenvx run -f /path/to/.env -- npx -y @lit-protocol/vincent-mcp-server http
 ```
 
 For an .env file example check [./.env.example](./.env.example)

packages/apps/mcp/package.json

@@ -29,6 +29,7 @@
   "main": "./dist/src/bin/cli.js",
   "scripts": {
     "dev:http": "tsx watch --tsconfig ./tsconfig.app.json --env-file=.env src/bin/http.ts",
+    "dev:stdio": "tsx watch --tsconfig ./tsconfig.app.json --env-file=.env src/bin/stdio.ts",
     "inspector": "npx @modelcontextprotocol/inspector",
     "integration:openAI": "tsx --env-file=.env ./integrations/openAI.ts",
     "integration:anthropic": "tsx --env-file=.env ./integrations/anthropic.ts"
@@ -44,12 +45,14 @@
     "express": "^5.1.0",
     "siwe": "^3.0.0",
     "tslib": "^2.8.1",
+    "which": "^5.0.0",
     "zod": "^3.25.64"
   },
   "devDependencies": {
     "@anthropic-ai/sdk": "^0.52.0",
     "@types/cors": "^2.8.19",
     "@types/express": "^5.0.1",
+    "@types/which": "^3.0.4",
     "openai": "^5.0.1",
     "tsx": "^4.19.4"
   }