Listen to this Post
Introduction:
The Model Context Protocol (MCP) is rapidly emerging as the de facto standard for how AI agents discover, authenticate, and execute against external tools and data sources. When Salesforce embedded a native MCP client into Agentforce, it unlocked a fundamentally new capability: AI agents that can reason on live external signals in real time — not just retrieve data that was synced hours ago. This distinction between “data availability” and “real-time reasoning” is reshaping how solutions architects design revenue technology stacks, particularly in customer success and retention use cases where signals change fast.
Learning Objectives:
- Understand the MCP client-server architecture and how Agentforce functions as an MCP host
- Differentiate between real-time reasoning use cases (MCP) versus data enrichment (APIs) versus orchestration (MuleSoft)
- Configure and secure an MCP server integration between Salesforce and external systems like Gainsight
- Implement OAuth 2.1 authentication and Zero Trust principles for Agentforce MCP deployments
- Apply Linux and Windows CLI commands for MCP server setup, testing, and monitoring
- Understanding the MCP Stack: Host, Client, and Server
The MCP architecture follows a straightforward client-server model where three components interact:
- Host: The AI application that facilitates interactions — Agentforce, Claude Desktop, VS Code, Cursor, or custom agents
- Client: Operates within the host to enable communication with MCP servers
- Server: Exposes specific capabilities — Tools (actions), Resources (data), and Prompts (reusable templates)
Agentforce includes a native MCP client in Pilot with the July 2025 release, enabling agents to connect to any MCP-compliant server securely without custom code. This means a Salesforce solutions architect can now configure Agentforce to pull live customer health scores from Gainsight, combine them with Salesforce CRM data, and make a judgment call in the moment — all without writing a single line of integration code.
Step-by-Step: Verify MCP Client Availability
Before configuring any MCP server, verify your Salesforce org supports the native MCP client:
Linux/macOS - Check Salesforce CLI version and MCP features sf version sf commands | grep mcp Windows PowerShell sf version sf commands | Select-String "mcp" Query active MCP servers via Salesforce DX sf data query -q "SELECT Id, Name, IsActive FROM MCPServerDefinition" -t
- The Real-Time Reasoning Use Case: Gainsight + Agentforce
Deron Tse, a Salesforce Solutions Architect, articulated a concrete example that crystallizes the MCP value proposition:
“Customer health scores, product usage signals, risk flags, and renewal timelines live in Gainsight. If an agent needs to reason on whether an account is at churn risk or ready for an upsell conversation, pulling that data live via MCP during the session makes more sense than waiting for it to appear in Salesforce through a nightly sync.”
This is the use case where MCP earns its complexity. The agent needs the data now, in context, to decide what to do next. Gainsight has officially joined the Salesforce Agentforce MCP beta, allowing customer-facing teams to view Gainsight intelligence within Salesforce and act on it through Salesforce’s AI tools.
Step-by-Step: Configure Gainsight MCP Server in Salesforce
- Create an External Client Application in Salesforce Setup → Applications → External Client Applications
2. Enable OAuth Settings with the following scopes:
- Access Salesforce Platform MCP services (
mcp_api) - Perform requests at any time (
refresh_token,offline_access)
- Activate the MCP Server in Setup → MCP Servers. Common built-ins include:
– `sobject-all` — full CRUD + SOQL + SOSL (9 tools, 2 prompts)
– `sobject-reads` — read-only subset
– `metadata-experts` — schema/describe helpers - Register the Gainsight MCP Server in Agentforce configuration
- Test the connection using the Salesforce DX MCP Server:
Linux/macOS - Install Salesforce MCP Server globally
npm install -g @kirtijha/salesforce-mcp-server
Configure MCP client (Claude Desktop example)
Edit ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"salesforce": {
"command": "npx",
"args": ["-y", "@kirtijha/salesforce-mcp-server"],
"env": {
"SF_ACCESS_TOKEN": "your_access_token",
"SF_INSTANCE_URL": "https://your-instance.salesforce.com"
}
}
}
}
Windows PowerShell - similar configuration in %APPDATA%\Claude\claude_desktop_config.json
- When MCP Is NOT the Right Tool: API Enrichment vs. Orchestration
Not every external system fits the real-time reasoning pattern. Tse makes a critical distinction:
- Intent data from Demandbase is most useful when it’s already in Salesforce as enrichment before a rep or agent acts on it — a well-configured API handles that cleanly
- Outreach activity from Salesloft triggering a complex downstream sales process involving multiple systems and approval workflows is an orchestration problem — that’s where MuleSoft fits better than MCP
The question every architect should ask: Does the Agentforce agent actually need to reason on external data in real time, or does the data just need to be available in Salesforce before the agent runs?
Step-by-Step: Decision Framework
Linux - Script to evaluate integration patterns !/bin/bash echo "Integration Pattern Decision Tree" echo "1. Does the agent need to reason on data in real-time?" echo "2. Is the data time-sensitive (churn risk, renewal signals)?" echo "3. Would a nightly sync introduce unacceptable latency?" If yes to all → MCP If data is static enrichment → REST API If multi-system orchestration → MuleSoft
4. Security-First MCP Deployment: Authentication and Zero Trust
MCP security operates at the transport layer with two supported transports:
- stdio — runs the MCP server as a local process; no authentication between client and server is required
- Streamable HTTP — uses OAuth 2.1 protocol for authentication and authorization
For enterprise deployments, the Streamable HTTP transport with OAuth 2.1 is mandatory. Agentforce is built to align with Zero Trust principles, which assume no user or system should be trusted by default.
Key Security Controls:
- OAuth 2.1 with PKCE (Proof Key for Code Exchange) for all MCP client-server communication
- Per-server authentication enforcement via MCP Security Gateway
- Cryptographic client identity verification using JWT and public keys
- Agent Variables — session-scoped values set only by trusted system actions
- Action Bindings — secure connections preventing unauthorized data injection
Step-by-Step: Configure OAuth 2.1 for MCP
Linux/macOS - Generate PKCE code verifier and challenge openssl rand -base64 32 | tr -d '\n' > code_verifier.txt echo -1 "$(cat code_verifier.txt)" | sha256sum | cut -d' ' -f1 | xxd -r -p | base64 | tr '+/' '-_' | tr -d '=' > code_challenge.txt Windows PowerShell $verifier = [bash]::ToBase64String([Security.Cryptography.RandomNumberGenerator]::GetBytes(32)) $verifier -replace '+','-' -replace '/','<em>' -replace '=','' | Out-File code_verifier.txt $sha256 = [System.Security.Cryptography.SHA256]::Create().ComputeHash([Text.Encoding]::ASCII.GetBytes($verifier)) [bash]::ToBase64String($sha256) -replace '+','-' -replace '/','</em>' -replace '=','' | Out-File code_challenge.txt Create Connected App in Salesforce with PKCE enabled Callback URL: https://your-app.com/oauth/callback Scopes: mcp_api, refresh_token, offline_access
5. Monitoring and Auditing MCP Traffic
Salesforce provides comprehensive observability for MCP interactions. Every tool call is logged and traceable. The MCP Security Gateway provides defense-in-depth interception for all MCP traffic.
Step-by-Step: Enable MCP Auditing
Linux - Query MCP audit logs via Salesforce CLI sf data query -q "SELECT Id, CreatedDate, MCPClientId, MCPServerId, Action, Status FROM MCPAuditEvent ORDER BY CreatedDate DESC LIMIT 100" -t Windows PowerShell sf data query -q "SELECT Id, CreatedDate, MCPClientId, MCPServerId, Action, Status FROM MCPAuditEvent ORDER BY CreatedDate DESC LIMIT 100" -t Monitor real-time MCP events sf data tail -o MCPAuditEvent
Key Metrics to Monitor:
- Authentication failures per MCP server
- Tool call latency (p50, p95, p99)
- Rate limiting breaches
- Unusual data access patterns
6. Deploying Custom MCP Servers
Beyond prebuilt servers, organizations can build custom MCP servers to expose proprietary APIs and data sources. The Salesforce DX MCP Server provides a developer-friendly foundation.
Step-by-Step: Build and Deploy a Custom MCP Server
Linux/macOS - Clone the MCP server template git clone https://github.com/salesforcecli/mcp cd mcp npm install Create a new MCP server npx @modelcontextprotocol/create-server my-mcp-server cd my-mcp-server npm install Define tools in src/index.ts Build the server npm run build Test locally with stdio transport node build/index.js Deploy to Salesforce Hosted MCP Server sf mcp deploy -1 "My Custom Server" -p ./build Windows PowerShell Similar steps, use cmd or PowerShell with Node.js installed
7. The Future: Unified Revenue Data Architecture
Tse’s third post takes this to a bigger canvas — unifying an entire revenue data architecture across the full lifecycle. As MCP matures, we’re seeing:
- AgentExchange as a marketplace for verified MCP servers
- Salesforce Hosted MCP Servers as a managed product
- MuleSoft MCP Bridge to quickly expose existing integrations as MCP servers
- Agentforce Context Protocol (ACP) — a Salesforce-1ative protocol inspired by MCP’s architecture
What Undercode Say:
- Key Takeaway 1: MCP transforms Agentforce from a data-retrieval tool into a real-time reasoning engine. The native MCP client in Agentforce 3.0 (July 2025) eliminates custom integration code for connecting to external systems. This shifts the architect’s role from writing integrations to designing which data should be accessible in real-time versus pre-synced.
-
Key Takeaway 2: The real-time reasoning vs. data availability distinction is the most critical architectural decision. Customer success and retention use cases (churn risk, renewal signals) demand MCP’s live data pull. Enrichment data (intent signals) and complex orchestration (multi-system workflows) are better served by APIs and MuleSoft respectively.
Analysis: The MCP ecosystem is evolving rapidly — from Anthropic’s 2024 draft to Salesforce’s native implementation in under 18 months. Organizations that adopt MCP early will gain a competitive advantage in AI-driven customer engagement. However, the security implications cannot be overstated. Agentforce agents inherit user permissions and operate with Zero Trust principles, but architects must rigorously enforce least privilege, monitor all MCP traffic, and regularly review access patterns. The distinction between MCP, REST APIs, and orchestration layers will become a standard part of every solutions architect’s decision framework. As Tse notes, “the question probably determines the right pattern more than the tool itself does”.
Prediction:
+1 MCP will become the default integration pattern for AI agents across all major enterprise platforms within 24 months, reducing custom integration costs by 60-70%.
+1 AgentExchange will evolve into a multi-billion dollar ecosystem of verified MCP servers, similar to the Salesforce AppExchange but specifically for AI agent capabilities.
-1 Organizations that fail to implement proper OAuth 2.1 authentication and Zero Trust controls for MCP deployments will face significant data exposure risks, with potential breach costs exceeding $4M per incident.
+1 The real-time reasoning capability will drive a new category of “adaptive customer success” where AI agents proactively intervene based on live signals, reducing churn by 15-25%.
-1 Legacy integration patterns (ETL, batch sync) will become obsolete for customer-facing AI applications, creating a skills gap for architects trained only in traditional API integrations.
+1 Salesforce’s investment in MCP and Agentforce positions them as the leader in enterprise AI agent orchestration, potentially capturing 40%+ of the emerging AI agent middleware market by 2028.
▶️ Related Video (84% Match):
🎯Let’s Practice For Free:
🎓 Live Courses & Certifications:
Join Undercode Academy for Verified Certifications
🚀 Request a Custom Project:
Secure, high-velocity infrastructure and disruptive technological engineering. Contact our engineering team for high-tier development and proprietary systems:
[email protected]
💎 Smart Architecture | 🛡️ Secure by Design | ⭐ Trusted by Thousands
IT/Security Reporter URL:
Reported By: Djtse Mcp – Hackers Feeds
Extra Hub: Undercode MoN
Basic Verification: Pass ✅
