ApexAgent
Documentation · Install · MCP server · Agent · Tools
💡 No extra Salesforce licensing. ApexAgent uses only standard platform features: Apex, custom objects, Lightning Web Components and a Salesforce Site. It needs no Agentforce license, no Flex Credits, no Einstein add-on and no Data Cloud. It runs in Developer, Enterprise, Performance and Unlimited Edition orgs.
The only running cost is the AI model itself, and only for the agent: you pay your model provider directly for what you use. The MCP server costs nothing to run, because the model belongs to the client that connects.
Everything deploys into your org and runs with your users' own permissions. There is no middleware and no external server to host.
ApexAgent gives you two things. They draw on the same set of tools, and you can use either one without the other.
| 🔌 MCP Server | 🤖 Agent |
|---|
| What it is | An endpoint in your org that lets an outside AI client (Claude, ChatGPT, or any MCP client) use Salesforce tools | A chat assistant inside Salesforce, in the utility bar |
| Where the AI model runs | In the client you connect. ApexAgent never calls a model. | Called from Apex, at any OpenAI-compatible provider |
| Extra Salesforce license | None | None |
| Who pays for the model | Whoever runs the client | You, through your account with the model provider |
| Needs a model provider API key | No | Yes. OpenRouter comes preconfigured, or use your own |
| Needs an External Client App, Salesforce's newer form of a Connected App | Yes. The deploy creates it | No |
| Users sign in with | Their own Salesforce login, from the client | Nothing extra; they are already in Salesforce |
Quick start
You need the Salesforce CLI, Git, and an org where you are a System Administrator. See the full prerequisites.
git clone https://github.com/badrisnarayanan/ApexAgent.git
cd ApexAgent
sf org login web --alias my-org --set-default
sf project deploy start --source-dir force-app
sf apex run --file seed/setup.apex
Setup creates a starter configuration: 2 LLMs, 10 tools, 1 agent (Lumen) and 1 MCP server. Then follow the install guide to give your users access and finish the part you want:
What's included
- Ten built-in tools for finding objects, reading and searching records, running SOQL queries and saving changes. Every tool that touches records runs as the signed-in user.
- ApexAgent Setup Portal: a Lightning app for managing LLMs, agents, topics, instructions, tools and MCP servers.
- Logs of every conversation, message, tool call and error, with reports and dashboards.
- A setup script that creates a working starter configuration.
Documentation
| |
|---|
| ApexAgent MCP server vs Salesforce Hosted MCP Servers | What each is, what each costs, when to choose which |
| ApexAgent agent vs Agentforce | The same comparison for the agent |
| Security and data | Who sees what, what reaches a model, and where data goes |
| Add your own tool | Write one Apex class and give it to an agent, an MCP server, or both |
| Use a different model | Switch the agent to another model |
| Use a different provider | OpenAI, Azure OpenAI, your AI gateway or any OpenAI-compatible service |
| MCP limits | The Salesforce limits that apply, and how to watch them |
| How it's organised | The records that make up an agent and an MCP server |
| Troubleshooting | Fixes for common setup and sign-in problems |
| Contributing | Repository layout, tests, and the starter configuration |
The documentation site's source is in site/.
License
MIT
