MCP Server for AI Development Agents

Stop wasting tokens on vague codebase searches. SalaMCP gives AI agents access to pre-indexed metadata structures and interdependencies from your Salesforce org—without ever sending your data anywhere.

Why This Exists

Working with AI agents on Salesforce projects is hard when they don't understand your org. They guess at object names, waste tokens searching through files, and constantly ask for context you should or you've already provided.

SalaMCP fixes this by running a local MCP server that indexes your Salesforce metadata. Your AI agent gets precise context about your org structure, and you burn through fewer tokens getting real work done.

Your Data Stays Local

Your Salesforce org authentication details or medata collection is never transmitted to, stored on, or processed by any servers affiliated with this extension

• Processing runs on your machine
• Uses your existing Salesforce CLI auth (the one you already have)
• Your org metadata never leaves your computer

Period.

What You Get

Better context for AI agents - They know what objects, fields, and relationships exist in your org. When you ask to add functionality, they understand your existing automation (triggers, flows, validation rules) and dependencies. No conflicts, no missing pieces.

Fewer wasted tokens - Precise queries mean less back-and-forth. Agent finds what it needs on the first try instead of guessing.

Reliable search - Queries structured metadata instead of fumbling with file paths.

MCP Capabilities

Tools:

• get_metadata_insights - Query your org's metadata structure

Resources:

• sObject descriptions for all standard and custom objects
• Order of Execution context per sObject

Sampling: When working with the agent, enable request sampling for more accurate suggestions tailored to your org's data patterns.

Beta Notice

This extension is in beta and has been tested on macOS only. Windows and Linux support is untested.

Setup

You need Salesforce CLI installed and authenticated to your org. If you can run sf org list and see your org, you're good.

After installing the extension:

1. Let it download the Node dependencies (happens automatically)
2. Open Command Palette and run: SalaMcp: Build Salesforce Org Metadata Index
3. Wait while it indexes your org

That's it. The index lives on your machine.

Using It

When you prompt your AI agent in 'Agent' mode, start with:

use salamcp tools

This tells the agent to use the MCP tools to understand your org.

Examples:

use salamcp tools - what fields are on the Account object?

use salamcp tools - find apex classes that use the Contact object

use salamcp tools - show me validation rules on Case

The agent queries your local index instead of fumbling around with file searches. Manually add resources, if needed. Its highly recommended.

Keeping It Current

Run the index build command again whenever you deploy new metadata or make structural changes to your org:

SalaMcp: Build Salesforce Org Metadata Index

The beta extension doesn't auto-update the index yet. You control when it syncs.

Technical Details

• Requires VS Code 1.80.0+
• Node.js 18.x+
• Salesforce CLI (any recent version)

When Things Break

Extension doesn't work after install? Check the Output panel in VS Code for errors. Most likely the Node dependencies didn't finish downloading.

Also check if:

• The MCP server is running (press CMD+SHIFT+P and execute "MCP: List Servers")
• The MCP server has been authorized in your agent session
• Sampling is enabled for Copilot LLM requests

AI agent ignoring the tools? Make sure you built the index first. Also check that your SF CLI is still authenticated (sf org list).

Index build fails? Your SF CLI auth probably expired or you don't have network access to your Salesforce org.

What This Isn't

This isn't a deployment tool, it's not a code generator, and it's not trying to replace your IDE's Salesforce extensions. It's a context provider for AI agents. That's all.

License

See LICENSE file.

Built because searching for Salesforce metadata by filename is terrible and AI agents deserve better context.