How to Build an OpenClaw Salesforce Agent for CRM Automation

Most Salesforce AI coverage focuses on Einstein AI or Agentforce, both locked inside Salesforce's own ecosystem. An OpenClaw Salesforce agent takes a different approach: it runs on your hardware, connects to Salesforce through MCP, and lets you query your CRM data conversationally from a terminal, chat interface, or messaging app.

The practical difference matters. Instead of switching to the Salesforce UI to pull pipeline numbers, you ask your agent: "Show me all opportunities closing this month over $50K." The agent translates that into a SOQL query, hits the Salesforce REST API, and returns structured results. You can then ask follow-up questions, request a summary grouped by stage, or have the agent create a report file you can share with your team.

Composio provides the managed MCP integration layer that handles OAuth tokens, API authentication, and scope management. The SFAILabs guide confirms that the core capabilities include SOQL query execution, opportunity tracking, lead management, and daily pipeline summaries. NVIDIA's NemoClaw announcement at GTC 2026 named Salesforce among its launch partners, validating the enterprise readiness of OpenClaw-based CRM workflows.

Salesforce serves over 150,000 customers globally, and 90% of Fortune 500 companies use the platform. Connecting an open-source agent framework to that data opens workflows that proprietary tools cannot match in flexibility.

Before connecting OpenClaw to Salesforce, you need three things: a working OpenClaw installation, a Salesforce org with API access enabled, and about 30 minutes.

Salesforce Connected App

Start in Salesforce Setup under App Manager and create a new Connected App with OAuth 2.0 settings. The SFAILabs guide walks through the full configuration: you'll set a callback URL, add OAuth scopes for API access and refresh tokens, and enable PKCE for security. Save and wait a few minutes for the Connected App to propagate.

Copy your Client ID and Client Secret. Store credentials in environment variables or a secrets manager, never in skill files or agent configurations. Credentials in workspace files risk exposure through logs or shared repositories.

Salesforce Instance URL

You also need your Salesforce instance URL. If your org uses a My Domain hostname (most do), that hostname is the instance URL. The connection authenticates through standard Salesforce OAuth and bearer tokens, so the same setup works whether you're on a Developer Edition, Sandbox, or Production org.

Security Note

Create a dedicated Salesforce integration user with a restricted permission profile. This user should have API access and read/write permissions only for the objects your agent needs. Test all operations in a sandbox org before pointing at production data. Field-level security gives you additional control over which data the agent can access.

Composio acts as the bridge between OpenClaw and Salesforce. It manages OAuth flows, token refresh, and API scoping so your agent doesn't need to handle authentication logic directly. There are two ways to set up the connection.

Plugin approach. Install the Composio plugin for OpenClaw, then configure your API key from the Composio dashboard. The plugin connects to Composio's MCP server and registers Salesforce tools directly into your agent. This is the more permanent setup and handles token refresh automatically. Refer to the Composio OpenClaw toolkit page for current installation steps, since commands and configuration keys may change between releases.

Setup prompt approach. For a quicker start, visit the Composio dashboard, copy the generated setup prompt, and paste it into your OpenClaw chat interface. Authenticate Salesforce through the dashboard, and you can start issuing commands immediately. This works well for testing before committing to the plugin route.

Both methods register the same Salesforce toolkit, which provides access to tools spanning Sales Cloud and Service Cloud. The most useful for CRM automation include:

• Contact and lead management: create, update, search, and score
• Opportunity handling: clone deals, add products, update stages
• Campaign automation: add or remove members, track engagement
• Task creation: link tasks to contacts, leads, or opportunities
• SOQL query execution: run arbitrary queries against your org
• Case management: create, assign, escalate, and close support cases

Your agent can call these tools from any OpenClaw interface, whether that's the terminal, a web UI, or a messaging integration like Telegram or Slack. Composio also lets you scope which Salesforce actions the agent can access, so you can disable write tools entirely for read-only reporting workflows.

Store your Salesforce reports where your whole team can find them

Free 50 GB workspace with semantic search, no credit card. Your OpenClaw agent writes CRM exports to Fastio via MCP, and your team accesses them instantly through the browser.

Start 14-Day Trial

The real power of an OpenClaw Salesforce agent shows up in conversational data access. Instead of building Salesforce reports through the UI, you describe what you need in plain language.

Lead Queries

Ask your agent to retrieve leads by status, source, or creation date. The underlying SOQL handles field selection automatically:

SELECT Name, Email, Company, LeadSource, Status
FROM Lead
WHERE Status = 'Open - Not Contacted'
AND CreatedDate = LAST_N_DAYS:30

The agent translates natural language requests like "Show me uncontacted leads from the past month" into queries like this, executes them, and formats the results.

Opportunity Tracking

Pipeline queries group results by stage for quick summaries:

SELECT StageName, COUNT(Id), SUM(Amount)
FROM Opportunity
WHERE IsClosed = false
GROUP BY StageName

This gives you a weighted pipeline view without opening Salesforce. You can drill down by asking follow-up questions: "Which of those closing this week have amounts over $25K?"

Contact and Account Lookups

Search contacts by email, find all contacts at a specific account, or filter by date ranges using Salesforce date literals like THIS_WEEK, LAST_MONTH, and LAST_N_DAYS:30.

Record Modifications

The agent can create new leads, update opportunity stages, and create tasks linked to specific records. The SFAILabs guide enforces confirmation requirements before any write operation, so your agent will ask for approval before modifying data. This is a smart safety mechanism for production orgs.

An OpenClaw agent that queries Salesforce generates artifacts: pipeline summaries, lead lists, deal reports, exported CSVs. Those files need to live somewhere accessible to your team, not buried in a local directory on whatever machine runs your agent.

Local filesystem storage works for personal use, but breaks down when multiple people need access or when agents run on ephemeral infrastructure. S3 or Google Drive can store files, but they lack built-in semantic search and require additional tooling to make files queryable.

Fastio provides workspaces designed for this workflow. Your OpenClaw agent writes CRM reports to a Fastio workspace using the MCP server, and anyone on the team can access them through a browser with previews, comments, and search. Enable Intelligence Mode on the workspace, and those reports become semantically searchable. Ask "What was our pipeline value last Tuesday?" and get an answer with citations from the stored reports.

The Business Trial includes 50 GB of storage, included credits, and 5 workspaces, with no credit card required. For a Salesforce automation workflow, that covers months of daily pipeline reports and lead exports.

The ownership transfer feature fits naturally here: an agent builds and maintains a reporting workspace, then transfers it to a sales manager who takes over as the primary owner. The agent retains admin access to keep updating reports, while the human controls sharing and permissions.

File locks prevent conflicts if you run multiple agents writing to the same workspace. Webhooks can trigger downstream workflows when new reports land, such as posting a Slack notification when the daily pipeline summary is ready.

Running an AI agent against production Salesforce data requires careful access controls. OpenClaw's local-first architecture helps here: your CRM data never passes through a third-party agent platform. Queries execute from your hardware, through Composio's authenticated MCP connection, directly to your Salesforce org.

Salesforce-Side Controls

• Create a dedicated integration user with a restricted profile. Grant API access and object permissions only for what the agent needs.
• Use field-level security to prevent the agent from accessing sensitive fields like Social Security numbers or financial details.
• Monitor agent activity through Salesforce's event logging and login history. Every API call is traceable to the integration user.
• Test all operations in a sandbox org first. Salesforce sandbox environments mirror production data without the risk.

OpenClaw-Side Controls

• The SFAILabs guide enforces confirmation rules for write operations. Keep these enabled in production so the agent asks before modifying records.
• Store credentials in environment variables or a secrets manager, never in skill files or agent configurations.
• NVIDIA's NemoClaw adds enterprise security controls on top of OpenClaw, including sandboxed skill execution and privacy guardrails. If your organization requires additional compliance layers, NemoClaw provides them without abandoning the OpenClaw ecosystem.

Composio Scoping

Configure which Salesforce actions your agent can access when setting up the Composio connection. If the agent only needs read access for reporting, disable write tools entirely at the Composio dashboard level. You can also bring your own OAuth credentials instead of using Composio's managed authentication, giving you full control over token lifecycle and scoping.