MCP tools
Reference the repo, env var, sandbox, automation, model, log, and agent-run tools exposed by the Mogplex MCP endpoint.
Mogplex exposes 29 user-scoped tools. The client loads their current JSON
schemas through tools/list, validates arguments locally when supported, and
sends each call to the same service layer used by the v1 API.
Discovery and sandboxes
| Tool | Scope | What it does |
|---|---|---|
mogplex_list_repos | read | List accessible repos and the GitHub installation id used by automations. |
mogplex_list_agents | read | List user-owned and preset agents available to Flow nodes. |
mogplex_list_skills | read | Search the user's skills. Returns handle, name, and description. |
mogplex_get_skill | read | Read one skill's full instructions by handle. |
mogplex_list_models | read | List enabled model ids. |
mogplex_list_sandboxes | read | List active or recent sandboxes, optionally by repo or status. |
mogplex_create_sandbox | write | Create or reuse a repo and branch sandbox. |
mogplex_get_sandbox_logs | read | Read install logs, dev-server logs, errors, and lifecycle events. |
A run prompt can invoke a skill without fetching it: write $handle in the
prompt of mogplex_start_agent_run. See
Using Skills.
mogplex_start_agent_run also takes mode (SAFE, AUTO, or YOLO) for
Codex and Claude Code runs. See Start a run for
what each mode allows. MCP tools aren't screened in SAFE and stay callable.
mogplex_create_sandbox consumes the existing launch event stream and returns
when the launch succeeds or fails. It does not create a separate status-polling
loop.
Creating a Sandbox through MCP does not create a Control Worktree. The public MCP catalog does not currently expose Worktree creation, Diff, Rebase, Archive, Prune, or worker-launch tools. Use the shipped Control Worktrees surface for those actions. See Control Execution Environments for the resource boundary.
Environment variables
| Tool | Scope | What it does |
|---|---|---|
mogplex_list_env_vars | read | List env var keys, targets, and types on the repo's linked Vercel project. |
mogplex_set_env_var | write | Create or update an env var by key, optionally scoped to targets. |
mogplex_delete_env_var | write | Delete every entry of an env var key. |
Env var tools operate on the Vercel project resolved for the repo. Listing
never returns values, so agents can reconcile keys without pulling secrets
into model context. When target is omitted, mogplex_set_env_var updates
the value of every existing entry with the same key, or creates one for all
targets (production, preview, development) when none exists. Passing
target when a key has multiple target-specific entries returns a CONFLICT
error — delete the key and recreate it instead. For repos whose
environment sync mode pulls from the linked Vercel
project, running sandboxes pick up changes on restart or resume; repos using
manual variables do not inherit these changes.
Automations
| Tool | Scope | What it does |
|---|---|---|
mogplex_list_automations | read | Return a bounded, cursor-paginated page of graph-free summaries. |
mogplex_get_automation | read | Read one automation with its complete draft and published graphs. |
mogplex_get_automation_schema | read | Read graph fields, validation rules, and scheduled-task examples. |
mogplex_validate_automation | read | Check a complete graph without saving or running it. |
mogplex_create_automation | write | Create a draft, optionally from a complete validated graph. |
mogplex_update_automation | write | Update metadata or replace the complete validated draft graph. |
mogplex_publish_automation | write | Validate, version, and activate the draft. |
mogplex_delete_automation | write | Permanently delete an automation that has no in-flight runs. |
mogplex_set_automation_model | write | Set or clear an agent node's model override. |
mogplex_trigger_automation | write | Run an active published automation for an owned repo. |
mogplex_cancel_automation_run | write | Cancel a Flow job using its automation ID and job run ID. |
mogplex_list_automation_runs | read | List recent executions and node or dispatch summaries. |
mogplex_get_automation_run_logs | read | Read node runs, dispatch events, AI calls, errors, usage, and review findings. |
mogplex_get_automation_run_artifact | read | Read a JSON file an API-triggered run committed under .mogplex/artifacts/. |
mogplex_rerun_pr_review | write | Queue another Mogplex PR Review run for a pull request. |
Automation listing defaults to 50 records and accepts a maximum limit of
100. Pass nextCursor from one response back as cursor. The list response
omits graphs; call mogplex_get_automation only when you need the full draft or
published graph.
Agent runs
| Tool | Scope | What it does |
|---|---|---|
mogplex_start_agent_run | write | Start a Codex, Claude Code, or Mogplex agent run in a repo sandbox. |
mogplex_get_run | read | Read current run state and metadata. |
mogplex_get_run_events | read | Read append-only run events. |
mogplex_cancel_run | write | Request cancellation of a one-off agent run, not a Flow job. |
Response shape
Successful calls include readable text for conversational clients and
structuredContent for clients that consume JSON directly. Tool errors return
the MCP error envelope with a stable machine-readable code such as
BAD_REQUEST, UNAUTHORIZED, FORBIDDEN, NOT_FOUND, or CONFLICT.
Call tools/list for the authoritative current JSON schemas. The
Mogplex MCP endpoint reference documents the
transport envelope and direct JSON-RPC request shape.