Tool surfaces
What each of the ten LEX AI mode surfaces exposes, the five tools present in every mode, and why a stale tool list returns a structured payload instead of an error.
Tool surfaces
Each mode exposes its own set of MCP tools. The unified server imports all ten surfaces at boot but mounts one at a time; switching modes swaps the mounted surface and refreshes the IDE tool list. Surfaces range from ten tools (Deploy) to seventeen (Test).
On every surface
Five tools appear in all ten modes and are omitted from the lists below: ask_user_question, record_user_answer, read_input_file, render_pdf_pages and switch_to_mode. See Files and questions.
Forward (and MVP Completion — same signatures)
Tool | Role |
|---|---|
| Create a NEW repo, |
| New run on an EXISTING repo — fresh |
| Rehydrate coordinator state after IDE restart/handoff. |
| Step brief + agent + docs; detects user edits and triggers re-execution. |
| Commit + push to the run branch; return next-step instructions. |
| Stage 1 audit template; stage 2 PR + squash-merge + close issue. |
| Diagnostics and progress snapshots. |
| Legacy pairwise switch. |
MVP Completion's get_plan_step refuses any step below 9.
Brief
kickstart_brief, get_next_question, submit_answer, skip_topic, open_interview_form, submit_form_answers, finish_interview_now, get_brief_status, finalize_brief. submit_answer accepts only the outstanding question id; finalize_brief names the build mode to hand off to.
Backward
reverse_kickstart, scan_project, generate_wiki, generate_questionnaire, submit_questionnaire, get_reverse_step, notify_reverse_complete, get_reverse_status, finalize_reverse, switch_to_forward_mode.
Edit
kickstart_edit, list_edit_tasks, propose_edit_plan, set_edit_plan, get_task_brief, notify_task_complete, waive_task, get_edit_status, finalize_edit. lex-compliance-check is never waivable, and migration-check is non-waivable once a model-shaped change requires it.
Review
kickstart_review, list_review_types, get_review_brief, notify_review_complete, get_review_status, finalize_review, switch_to_forward_mode. Built-in types: convention, business.
Test
kickstart_test_run, list_responsibilities, propose_test_plan, set_test_plan, get_responsibility_brief, notify_responsibility_complete, waive_responsibility, record_scenario_matrix, record_test_execution, dispatch_reverse_prerequisite, get_test_status, finalize_test_run. Five of the ten responsibilities cannot be waived, and dispatch_reverse_prerequisite exists because the session will not open without a user-story document to test against.
Input
kickstart_input_change, register_format_change, get_intake_brief, list_change_areas, get_change_brief, notify_change_complete, get_input_change_status, finalize_input_change. notify_change_complete refuses a completion whose declared artifacts are not on disk.
Deploy
kickstart_deployment, get_deployment_step, notify_deployment_step_complete, get_deployment_status, finalize_deployment. The smallest surface, because the step bodies come from the remote Lex deployment handbook rather than a local catalogue.
MVP Generator
kickstart_mvp, list_mvp_checkpoints, submit_mvp_plan, submit_mvp_scope, scaffold_mvp, record_implementation, run_boot_checklist, get_mvp_status, finalize_mvp.
Cross-mode: the universal switch
switch_to_mode(target_mode, confirm_mode_switch, user_request, ...) flips the mounted surface, reloads .env, writes a one-shot override marker, and sends a burst of tool-list-changed notifications so the IDE refreshes. The legacy pairwise helpers remain on the surfaces that always carried them; prefer switch_to_mode in new work.
Why you may see a "tool not found" message
If the IDE cached a previous mode's tools, calling a tool from the old surface returns a structured stale_tool_call: true payload rather than a raw method-not-found error. The fix is always the same: refresh the tool list, switch mode if a suggested_mode is named, then retry — never work around it. See Mode switching.
Was this helpful?
Your feedback shapes what we write next.
