Arca MCP Server

Model Context Protocol (MCP) server implemented in TypeScript for the Arca project.

Structure

• src/tools/: Tools exposed by the server.

• src/commands/: Business logic and commands.

• src/prompts/: Prompt templates and context.

• src/vector_store/: Embedding, chunking and persistence logic with PostgreSQL and pgvector.

Related MCP server: ArcaMCP

Capabilities

Available Tools

• query_arca_docs: Semantic search (RAG) in AFIP technical documentation.

• get_ta: Obtains and caches the AFIP WSAA Access Ticket (Token and Sign) using your .crt and .key certificate.

• get_ultimo_comprobante: Queries the last authorized voucher number in AFIP for a point of sale.

• create_invoice: Generates and authorizes an electronic invoice (A, B or C) directly in AFIP, auto-calculating VAT amounts.

Available Resources

• resource://arca/guides/wsaa: Step-by-step guide for AFIP authentication.

• resource://arca/guides/invoice/{type}: Guide for issuing specific vouchers (A, B, C, MTXCA).

Requirements

• Node.js (v18+)

• Yarn

• PostgreSQL with the pgvector extension enabled.

Installation

yarn install

Usage

For development in IDEs/CLI (via Stdio):

yarn dev

Local Development (HTTP and MCP Inspector)

If you want to test the server locally over the network and use the web Inspector to debug the tools, follow these steps:

1. Start the server in HTTP (Streamable) mode:

   yarn run dev:http

   The server will listen on http://localhost:3001/mcp.

2. Start the MCP Inspector: Open a new terminal in the project and run:

   yarn run inspect

   This will open a window in your browser.

3. Connect the Inspector to the local server: In the Inspector web interface, if it allows you to configure HTTP Streamable transport, point it to http://localhost:3001/mcp. Otherwise, you can still use it in standard mode (Stdio) by running the Inspector pointing to the local index directly (npx @modelcontextprotocol/inspector node dist/index.js). Click "Connect" and you will see all the tools exposed by the server listed and ready to be tested.

To compile and run in production:

yarn build
yarn start