Resume work in another agent
Move from Codex to Claude or Cursor. The next agent receives the current decisions, completed work, open risks and source evidence instead of reconstructing the project from files and chat history.
Agent integration
Continue the same project in another agent.
Connect Codex, Claude, Cursor, OpenClaw or a custom runtime to persistent project memory. The agent receives the decisions, completed work, evidence and access boundary relevant to its task.
Updated July 30, 2026
What changes
The agent starts from project state, not from an isolated chat.
SPM keeps project knowledge outside any single agent. Connected tools can recover it, use it within their authorized scope and return the outcome for the next task.
Investigate an earlier decision
Ask why an API constraint still applies. The agent can move from a compact project view to the approving source, related evidence and superseded history when the answer needs detail.
Prepare a restricted handoff
Give a partner the integration contract and checklist while keeping pricing, private discussion and unrelated work outside the shared view.
Setup
Authorize once, then work through the agent.
- 01
Sign in and authorize
Connect the agent to your SPM account through the browser authorization flow.
- 02
Choose the project
Associate the task with an existing project, create a new one or continue without SPM memory.
- 03
Work naturally
The connector retrieves useful project knowledge and reports completed work back to the same project.
Supported paths
Use the same memory service from different agent tools.
Codex
Install the SPM connector and authorize it from the browser.
Claude and desktop MCP clients
Use the hosted MCP endpoint with the same project access model.
Cursor and IDE agents
Connect local project work to SPM through MCP or the API.
Custom runtimes
Use the SDK, CLI or API when the agent lifecycle is controlled by your application.
Technical connector boundary
SPM exposes recall, project resolution, context composition, controlled sharing, policy preflight and post-action reporting through MCP. Administrative billing and destructive controls remain outside the agent surface.
Read-only, daily-work and hardening profiles let operators choose the smallest tool surface appropriate for an agent run. The SDK, CLI and API use the same project and authorization contracts.