Give your Salesforce agents a way back.
bluefactory lets agents save a checkpoint before changing Salesforce, investigate what changed, and prepare a recovery you approve.
Agent tools in development. Snapshots, point-in-time query, and previewed restore ship today.
From agent action to approved recovery
Follow one renewal update, from saving the original records to undoing a mistake.
“Close the expired renewals.”
Renewal Agent
Acme renewal
and any MCP clientBefore the write
Give the agent a way back.
The agent asks bluefactory to save the records it is about to change, including the relationships needed to restore them.
create_checkpoint()CHK-3921 ready · 382 records saved
Bring protection to the assistant you use
Connect your AI assistant to bluefactory’s backup, history, and recovery tools through our MCP server.
Your administrator chooses the Salesforce orgs and tools each agent can access. Custom agents and automations can use the same tools through the REST API.
Designed for MCP-compatible assistants
AgentforceChatGPTClaudeIncluding internal agents and custom MCP clients. The hosted MCP server and REST API are in development with design partners.
Explore the MCP tools7 tools
Planned MCP endpoint
https://mcp.bluefactory.io/mcpGrant tools per org. Keep production recovery behind a named person's approval.
- create_backup
- Start a backup when an operation needs broader protection.Scope: backup:create
- create_checkpoint
- Save the records an agent will change and their related records.Scope: checkpoint:create
- query_backup
- Find previous values or deleted records in a captured snapshot.Scope: data:query
- get_record_history
- Read a record’s captured versions and what changed between them.Scope: history:read
- prepare_restore
- Preview a recovery and its conflicts before anything is written.Scope: restore:preview
- execute_restore
- Apply an approved recovery plan. Requires separate permission.Scope: restore:execute
- get_operation_status
- Check whether a backup, checkpoint, or restore is ready.Scope: backup:read
Set the boundaries. Approve the recovery.
Scope each agent’s access
Choose the orgs, objects, and tools an agent can use. Reading history and writing a restore are separate permissions.
Approve before restoring
Let the agent investigate and prepare the plan. Require a named person to approve the recovery before it changes production.
Preserve later human edits
Reverse the agent’s fields while keeping later edits to other fields. Changes to the same field are flagged as conflicts for review.
A few practical questions
What is available today?
Scheduled snapshots, point-in-time query, previewed restore, and metadata change tracking ship today. The MCP server, public REST API, agent checkpoints, and conflict-aware rollback are in development with design partners.
How will we connect our agents?
An administrator connects Salesforce through OAuth, configures the MCP connection in a compatible assistant, and grants access to the relevant orgs and tools. The agent’s workflow must call bluefactory and wait for its checkpoint to be ready before writing. Custom integrations can use REST.
Does bluefactory sit between the agent and Salesforce?
The agent writes directly to Salesforce. It calls bluefactory to create a checkpoint, read captured history, or prepare a recovery. bluefactory does not proxy the agent’s Salesforce traffic.
What if an agent skips the checkpoint?
Scheduled snapshots still provide recovery points, but they may not contain the state immediately before the agent’s change. Connecting MCP alone does not enforce a checkpoint. Your agent’s execution workflow must require it before writing.
Where does the saved data live?
In bluefactory-managed storage, or in your own Amazon S3 or Azure Blob bucket with your account, region, and keys. Captured data is encrypted at rest, with tenant isolation on reads.
Put a recovery plan behind your agents.
We are working with teams putting agents in front of production Salesforce data. Join us as a design partner.