The engagements we run most often on Anthropic Claude, from first implementation through optimization.
Evaluate your CRM data quality, security posture, and compliance requirements to ensure your Salesforce or HubSpot environment is prepared for safe Claude deployment.
Connect Claude directly into your Salesforce or HubSpot workflows to automate client communications, summarize records, and generate compliant content at scale.
Build audit-ready approval processes and field-level controls that embed regulatory requirements directly into your Claude-powered automation.
Equip advisors and relationship managers with Claude-powered tools that draft emails, prepare meeting summaries, and surface insights.
Establish data governance frameworks and quality standards that keep Claude operating on clean, accurate CRM data with audit trails.
Continuous monitoring, prompt refinement, and performance tuning so your Claude implementations keep delivering measurable business value.
Engagements are measured by movement on the numbers that matter. These are the directions of travel we commit to.
Predictable phases. Clear deliverables. No surprises.
One to two working sessions to map your current state, business goals, and gaps. We come out with a written scope and recommendation.
Documented architecture, realistic timeline, and transparent commercial proposal. No surprises and no hidden scope.
Configuration, development, integrations, data migration, and QA, with weekly demos and on-the-fly adjustments.
Training, change management, hypercare, and ongoing optimization. We do not disappear at go-live.
Consultant-level analysis from the consultants delivering the work.
How Anthropic Claude works with Salesforce alongside Agentforce: the ways the two connect, the jobs Claude does well on CRM data, and the security and data-handling decisions to make before you connect them.
Read the full post →Developers call Claude through the Anthropic API, or through cloud platforms such as Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry. Tool use lets Claude request a defined function, such as querying Salesforce accounts or creating a task, which your code executes. Your application decides which Salesforce calls exist, so Claude can only request actions you have explicitly built.
Salesforce offers hosted Model Context Protocol servers that let MCP clients such as Claude reach org data and actions through an external client app and OAuth. Access runs as the signed-in user, so field-level security and sharing still apply. Salesforce provides separate read-focused and write-capable server configurations, which makes a read-only pilot straightforward to set up first.
Salesforce and Anthropic announced in 2026 that Claude models are available to Agentforce through Amazon Bedrock inside the Salesforce trust boundary. Einstein Trust Layer controls, such as masking and audit, apply to those interactions. This path suits teams that want Claude’s reasoning while keeping agent design, subagents and actions in Salesforce. Confirm current availability for your org and edition with Salesforce.
Claude handles long inputs such as contracts, case histories, call transcripts and product manuals in one request. For Salesforce teams, that supports summarizing a year of case activity before a renewal, comparing a redlined contract with opportunity terms or drafting a handoff note. Output should land in a draft field or record for a person to review before it becomes official.
By defining a JSON schema for a tool or response, you can ask Claude for fields such as case category, sentiment and next step. The results arrive in a predictable shape. Your integration then validates values against Salesforce picklists and required fields before writing anything. That validation layer is what separates a reliable enrichment process from free text pasted into records.
Claude Code is Anthropic’s agentic coding tool, and Salesforce teams use it to read metadata, draft Apex classes, Lightning web components, test classes and deployment scripts. It works best inside a source-tracked project with the Salesforce CLI and sandbox orgs. Every change should still pass code review, automated tests and your normal deployment pipeline before reaching production.
Choose two or three tasks with clear value, such as account briefs or case summaries, and list exactly which objects and fields each needs. Classify that data with your security team, including regulated or personal information. Decide which Claude surface is acceptable for it before anyone connects an assistant, so the boundary is agreed rather than discovered.
Match the path to the use case. Hosted MCP servers suit assistants working as signed-in users, the API with tool use suits custom applications and Agentforce suits agents inside Salesforce. Create an external client app or integration user with only the scopes each path requires. Document which path handles which task so ownership stays clear.
Launch with read access only, and collect real questions and records into an evaluation set with expected answers. Score responses for accuracy, completeness and whether Claude declined appropriately when data was missing. Rerun the same set after every prompt, model or permission change, so improvements are measured rather than judged from a few impressive demos.
Introduce write actions one at a time, such as creating a task or drafting a case comment, with a person approving each change before it saves. Validate values against picklists and validation rules in code. Log the prompt, tool call, approver and resulting record ID, so any update can be traced and reversed if needed.
Review a sample of conversations and writes each week, track evaluation scores and watch for new failure patterns as users try unexpected requests. Retire prompts nobody uses and tighten tool descriptions that cause confusion. Only widen access to more objects, users or autonomous actions once the evidence shows the current scope behaves reliably.
Claude is available through Anthropic’s Team and Enterprise plans and through the Claude API, which is billed on token usage. It is also offered on Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry. Using Claude inside Agentforce is contracted through Salesforce, so confirm how that consumption is metered with your account executive. Salesforce says hosted MCP servers are included in Enterprise Edition and above, but verify the terms for your org. On data handling, Anthropic’s commercial terms state it may not train models on customer content, and zero data retention is available on request. Policies change, so review the current terms at anthropic.com and your cloud provider’s agreement before connecting production data.
Wiring Claude to Salesforce through a system administrator account exposes every record and every action to any prompt. A single misread instruction could update hundreds of records. Use per-user OAuth where possible, or a narrowly scoped integration user, and grant only the objects, fields and actions that each approved use case actually requires.
Emails, web-to-case submissions and chat transcripts are written by outsiders, and a message can contain text designed to redirect an AI assistant. Treat record content as untrusted input. Keep write actions behind approval, restrict which tools a conversation can reach and test your setup with deliberately adversarial cases before launch.
A few impressive answers in a workshop say little about performance on messy real records. Without an evaluation set, teams cannot tell whether a prompt change helped or hurt. Collect representative questions with known correct answers, score them before launch and rerun them after every model, prompt or permission change.
When summaries or classifications write straight into official fields, errors spread into reports, forecasts and customer communications without anyone noticing. Store AI output in clearly labeled draft fields or related records until a person confirms it. Keep the audit trail, so reviewers can see what Claude proposed versus what a person approved.
Agentforce fits agents that live inside Salesforce channels and rely on Salesforce topics, actions and the Einstein Trust Layer. Claude on its own fits cross-tool assistants, document-heavy work, coding and custom applications built outside Salesforce. Many organizations use both: Claude as a reasoning model within Agentforce where supported, and Claude connected through MCP for employees who work across Salesforce, Slack and other systems.
Anthropic’s commercial terms state that it may not train models on customer content from its commercial services, which cover the API and business plans. Zero data retention arrangements are available on request for eligible API features. Consumer accounts follow different terms, so staff should not paste customer data into personal accounts. Confirm the current policies at anthropic.com and with your cloud provider, since terms can change.
It depends on how you connect. Salesforce hosted MCP servers authenticate as the signed-in user, so field-level security, sharing rules and permissions apply as they would in the Salesforce interface. A custom integration using a single integration user sees whatever that user sees, which may be far more than the person asking. Prefer per-user authentication for assistants that answer employee questions.
Start read-only, then add writes one at a time with a person approving each change before it saves. Define tools narrowly, such as create follow-up task rather than a general update-any-record action. Validate values in code against picklists and required fields, log every call with its approver and test with adversarial inputs. Autonomy can expand gradually once evidence shows the scope is reliable.
We are a Select Services Partner (Claude Partner Network). In practice, our work focuses on the Salesforce side of these projects. That means choosing connection paths, scoping integration users and permissions, and building evaluation sets. It also covers approval steps for write actions and deciding where Agentforce, Claude or both belong. Product roadmaps, model availability and contract terms remain Anthropic’s and Salesforce’s to confirm directly.
Make sure your CRM is ready for Claude before you deploy, data, security, and compliance.
Connect Claude into Salesforce and HubSpot to automate comms, summarize records, and draft content.
Audit-ready approval processes and field-level controls embedded into Claude automation.
Claude-powered tools that draft emails, prep meeting summaries, and surface client insights.
Clean, governed data is the difference between an AI agent you can trust and one that confidently gives wrong answers.
Continuous monitoring, prompt refinement, and tuning so Claude keeps delivering value.
Tech Talk
A monthly brief for the people who own Salesforce, AI and revenue technology
What changed in Salesforce and AI this month, and what to do about it.
One email a month. Written by the consultants who deliver the work, not by a marketing team, for the leaders who make the technology decisions.
Consultant analysis, not vendor recaps. One click to leave.
