Atlas sees MCP servers as the quickest way to ship agent-safe automations. Drop our hosted server URL into any client, or fork the repo to extend tools. Either way you keep a single ledger for roles, scopes, provenance, and usage.
The short definition
A Model Context Protocol (MCP) server is a typed tool catalog that any compliant client can call over a simple JSON transport. Instead of wiring prompts to brittle HTTP endpoints, you describe each tool—inputs, outputs, scopes, cost hints, and rate caps—and publish the server URL. Clients discover the catalog, negotiate auth, and invoke tools with full context about latency, budgets, and consent requirements.
Atlas Agent treats the MCP server as the control plane for real estate automation. We preload two production tools (generate sales comps PDF and generate rental comps PDF) plus the ledger that tracks identity, consent, provenance, and usage. Teams start with a working server, then add tools from the marketplace or build their own with the same contract.
Why teams use MCP servers
Teams adopt MCP because they want predictable automation that plugs into any bot. The protocol separates “what the tool does” from “where the chat surface lives”, so you can switch between Custom GPTs, Slack bots, or the hosted Atlas Bot without rewriting integrations.
- Standardize prompts and tool behaviors so ops, agents, and vendors stay aligned.
- Keep identity, consent, provenance, and budgets in a shared ledger instead of shadow worksheets.
- Unlock partner tools: once a vendor posts an MCP endpoint, every client or bot in your stack can call it.
How it works at a glance
An MCP deployment has three roles: host, client, and server. You can host Atlas on our managed infrastructure or self-host anywhere that can serve HTTPS. Clients are the chat surfaces—Custom GPT, VS Code, Slack, CRM chat bars—that know how to discover a server catalog. The server advertises tools, resources, prompts, and metadata like rate caps.
When a user triggers a tool, the client packages parameters, identity claims, and consent scopes. The server preflights the request by reading policy counters, then executes or queues the tool. Postflight, Atlas signs the usage event with latency, vendor, cost, consent scopes, and artifact URIs so you can audit it later.
Atlas approach to MCP
Atlas Agent layers a knowledge base on top of the open MCP spec so buyers and developers understand the guardrails. Product pillars cover the full lifecycle: the tool library, tracker & guard, identity & consent, provenance, local booster, and the marketplace. Every section links back to the Knowledge Base so your security and legal teams see exactly how roles, scopes, and schemas behave.
Compare: MCP vs. RAG
Retrieval-augmented generation (RAG) pulls in unstructured context for a single response. MCP servers expose deterministic tools with typed outputs, budgets, and audit trails. Most enterprise assistants run both: they call MCP tools to fetch structured data (such as comps PDFs) and pair those artifacts with retrieval for narrative answers.
Atlas documents how the two models reinforce each other in /mcp-vs-rag. We also cover how provenance tags feed downstream RAG systems so they only cite artifacts that passed consent and tracker checks.
Build one today
You can deploy the Atlas default server, fork it, or build net-new MCP servers. The tutorials show both paths: how to compose a Python server with the Atlas tool contract, how to wire VS Code or another IDE for local testing, and how to connect your bots once the server is live.
Every guide links back to pricing so you know when to upgrade from Server Builder to Bot Starter Kits or Hosted Atlas Bot seats.