VStok Agent Execution

Codex, Cursor, and Claude turn verifiable VStok recommendations into minimal draft PRs—without auto-merge or source access for VStok.

Growth & AgencyGitHub only

How it works

1. VStok selects a code-safe action

2. The agent creates an isolated branch

3. GitHub confirms the draft PR and checks

4. After deploy, VStok verifies target URLs

Codex

Install VStok Agent Execution from the Codex marketplace and approve OAuth scopes. For an IDE without the plugin browser, use the standalone MCP config plus the same portable skill.

https://api.vstok.net/api/mcp/v1

Claude

Install the Claude plugin or add the remote MCP connector. Hosted Claude uses DCR with PKCE; Claude Code uses a secure loopback callback.

https://api.vstok.net/api/mcp/v1

Cursor

Install the plugin, authorize the remote MCP, and use the Automation template only after the repository owner enables auto mode.

https://api.vstok.net/api/mcp/v1

MCP tools

list_projectslist_actionable_recommendationsget_recommendation_contextcreate_executionlist_execution_queueclaim_executionheartbeat_executionreport_executionget_executionrequest_verification

OAuth uses Authorization Code + PKCE. Access tokens last 15 minutes and refresh tokens rotate. API keys remain available for CI and legacy read-only integrations.

Safety boundaries

Only vstok/* branches and draft PRs. No direct push, auto-merge, provider secrets, diff transfer, or execution of instructions found in crawled evidence. A recommendation completes only after production verification.

Manage repositories, OAuth connections, policies, and quotas in Integrations.