"namespace:io.github.smq9sn5jck-cloud" matching MCP tools:

• replay_sandbox_test

  Keploy

  Replay the sandbox test for one or more suites against captured mocks — re-runs the suite's steps against the dev's locally-running app while keploy serves outbound calls (DB, downstream HTTP, etc.) from the captured mocks. Use this when the dev says "replay", "run my sandbox tests", "integration-test", "check if mocks still match" — keywords "sandbox" / "replay" / "mocks" / "integration-test" all map here. Also the REPLAY STEP of FROM-SCRATCH: call this LAST (after create_test_suite + record_sandbox_test) to give the dev the whole-app regression picture against the freshly captured mocks. Output produces a SANDBOX RUN REPORT — it answers "does the suite still hold up against its captured baseline?". ═══════════════════════════════════════════════════════════════════ DISAMBIGUATION — pick this tool vs. replay_test_suite: ═══════════════════════════════════════════════════════════════════ USE replay_sandbox_test (THIS TOOL) when the dev says: * "run my sandbox tests" / "replay my sandbox tests" * "integration-test my app" / "run the integration tests" * "check if my mocks still match" / "replay against the captured mocks" * "rerun my sandbox suite" (with the word "sandbox") Trigger keyword: an explicit "sandbox" / "replay" / "mocks" / "integration-test" — silent signal that the dev wants captured-mock replay, NOT live-app execution. USE replay_test_suite INSTEAD when the dev says: * "run the test suite" / "run my test suites" (bare — no "sandbox") * "execute test suite X" / "run suite 810d3ebe…" * "test the suite again" / "smoke test against the live app" Bare verbs ("run / test / execute") applied to "the suite" without the word "sandbox" mean LIVE-APP execution, NOT captured-mock replay. replay_test_suite hits the dev's running localhost app directly via HTTP — no docker spin-up, no mocks. After a record_sandbox_test run, the natural next step is THIS tool (replay against the just-captured mocks). After create_test_suite / update_test_suite, the natural next step is replay_test_suite (validate against the live app). When the dev's verb is bare and the prior turn doesn't make the intent obvious, ASK rather than picking sandbox-replay silently — code-change regressions can hide under "mock didn't match" failures. ═══════════════════════════════════════════════════════════════════ DISCOVERY — when the dev hands you a bare suite_id with no app_id / branch_id: ═══════════════════════════════════════════════════════════════════ Suites live on a (app_id, branch_id) tuple. A bare suite_id has NO on-disk hint about which app or branch holds it; you have to RESOLVE both before calling this tool. Walk these steps in order — STOP as soon as getTestSuite returns 200: 1. Detect the dev's git branch: Bash `git rev-parse --abbrev-ref HEAD` in app_dir. If exit non-zero / output is "HEAD" → not a git repo / detached HEAD; ASK the dev for the Keploy branch name. 2. Resolve candidate apps via the cwd basename: Bash `basename $(pwd)` → call listApps with q=<basename>. Usually 1–2 candidates. If 0 → ASK; if >1 → walk every candidate in step 4. 3. For each candidate app, call list_branches({app_id}) and find the branch whose `name` matches the git branch from step 1. That gives you {branch_id}. If no match → not this app, try next. 4. Verify with getTestSuite({app_id, suite_id, branch_id=<from step 3>}). 200 → resolved; 404 → wrong app/branch, try next. 5. If steps 2–4 exhaust, walk every OPEN branch on each candidate app via list_branches → getTestSuite. Then try main (branch_id omitted). If still nothing → ASK the dev for the {app_id, branch_id} pair. After resolving once in a session, REUSE the {app_id, branch_id} for subsequent suite-targeted calls; don't re-walk discovery for every action. SCOPE — whole-app vs single-suite: * Default: LEAVE suite_ids UNSET → the tool resolves "every suite for the app that has a sandbox test (test_set_id populated)" and replays them all. Use this for "run my sandbox tests" / "check if my tests still pass" — whole-app regression. New suites auto-pick up. * Single / subset: PASS suite_ids when the dev names specific suites — "replay sandbox test for suite 810d3ebe-…", "replay only the auth suite", "run suite X and Y". The tool validates each requested id is actually a suite with a sandbox test (has test_set_id); an unlinked id gets a precise "record first" error instead of an opaque downstream CLI failure. This tool resolves the app, picks the suite set per the rule above, and returns a single playbook that drives the replay for them. It does NOT record. WHAT THIS TOOL DOES INTERNALLY (so you don't have to): 1. Resolves app_id — use the explicit app_id if the caller has one; otherwise pass app_name_hint (usually the cwd basename) and the server does listApps with a substring match. Multiple matches → error listing them; zero matches → error suggesting the dev generate a suite first. 2. Lists test suites for the app, keeps only those with a non-empty test_set_id. Zero linked → typed "no linked sandbox tests" error. 3. If suite_ids was passed, validates every requested id is in the linked-suites set; unlinked ids → typed error pointing to record_sandbox_test. 4. Returns the headless playbook — walk it exactly: spawn CLI in background, tail the progress file (PID-alive guard built in), read the terminal event, fetch the report. No separate cleanup step — the CLI exits on its own. ===== PREREQUISITES ===== (Same as record_sandbox_test — if you just recorded, you already have them. Same docker-compose network rule applies: use the same compose file + service, stop the app service before calling, leave deps running.) - app_command: shell command that starts the dev's app (e.g. "docker compose up producer"). - app_url: base URL the app listens on, e.g. http://localhost:8080. - app_dir: absolute path to repo root. - container_name if app_command is docker-compose. - keploy binary on PATH. If `which keploy` returns nothing, install it before calling this tool with: `curl --silent -O -L https://keploy.io/install.sh && source install.sh`. ===== AFTER CALLING — walk the playbook ===== Same headless playbook shape as record_sandbox_test: spawn `keploy test sandbox --cloud-app-id …` in the background via Bash, poll `tail -n 1 $PROGRESS_FILE` repeatedly (no sleep loops; the wait_for_done step has a built-in `kill -0 $KEPLOY_PID` guard so the loop exits if the CLI dies silently), read the terminal NDJSON event (phase=done, data.ok, data.test_run_id), and — if ok=true — call get_session_report(app_id, test_run_id) with verbose=true at the end. No separate cleanup step needed; the CLI exits cleanly once phase=done is written. ===== MANDATORY OUTPUT — Phase 3 section ===== Your final message to the dev MUST contain a section with this exact heading (do NOT merge with Phase 2; do NOT compress the failed-steps table even when failures are homogeneous): ### Phase 3 — Sandbox run report Under it, emit the uniform three-subsection format owned by get_session_report: (i) per-suite table — one row per suite in per_suite, passing suites included, columns = Suite name | passed/total steps. (ii) failed-steps table — ONE ROW per entry in failed_steps[], columns = Suite | Step name | Method + URL | Expected → Actual status | mock_mismatch y/n. Never collapse rows. (iii) Diagnosis + Recommendation (see get_session_report description for case-specific rules around mock_mismatch_dominant, repo-diff inspection, and the SKIP / FIX-CODE / FIX-TEST branching for fix-it follow-ups). Do NOT print aggregate step totals across suites — they mix unrelated suites and hide where damage actually is. ===== ROLLUP LINE ===== Close the message with a final one-line rollup paragraph (no heading), in addition to the three phase sections. Mention the TOTAL number of suites replayed (which may exceed the count created in this session, because replay_sandbox_test covers every linked suite the app has). Example: "_Rollup: inserted 4 suites, 4/4 with sandbox tests after record, 3/4 suites passed sandbox replay across the app's 6 linked suites — 1 failure is likely keploy egress-hook, file an issue with the IDs above._" ===== DO NOT ===== * DO NOT call update_test_suite or record_sandbox_test after this. The dev said RUN, not REFRESH. * DO NOT fall back to raw keploy CLI (`keploy test …`) if the MCP tool drops mid-flow — CLI runs test-sets directly and does NOT write results back to the MCP-visible TestSuiteRun. See MCP DISCONNECT RECOVERY in the top-level instructions.

  Connector

• create_snapshot

  BorealHost

  Create a local container snapshot (async). Runs in background — returns immediately with status "creating". Poll list_snapshots() to check when status becomes "completed" or "failed". Available for VPS, dedicated, and cloud plans (any plan with max_snapshots > 0). Local snapshots are stored on the host disk and count against disk quota. Requires: API key with write scope. Args: slug: Site identifier description: Optional description (max 200 chars) Returns: {"id": "uuid", "name": "snap-...", "status": "creating", "storage_type": "local", "message": "Snapshot started. Poll list_snapshots() to check status."} Errors: VALIDATION_ERROR: Max snapshots reached or insufficient disk quota

  Connector

• create_b2_snapshot

  BorealHost

  Create a B2 cloud-backed snapshot (zero local disk, async). Streams container data directly to Backblaze B2 via restic. No local disk impact — billed separately at cost+5%. Runs in background — returns immediately with status "creating". Poll list_snapshots() to check when status becomes "completed". Only available for VPS plans. Requires: API key with write scope. Args: slug: Site identifier description: Optional description (max 200 chars) Returns: {"id": "uuid", "name": "...", "status": "creating", "storage_type": "b2", "message": "B2 cloud snapshot started. Poll list_snapshots()..."} Errors: VALIDATION_ERROR: Not a VPS plan or max snapshots reached

  Connector

• upload_media

  SendIt

  FOR CLAUDE DESKTOP ONLY (with filesystem access). For Claude.ai/web: Use create_upload_session instead - it provides a browser upload link. Upload local media to cloud storage, returning a public HTTPS URL. WHEN TO USE: • Instagram, LinkedIn, Threads, X: REQUIRED for local files before calling publish_content • TikTok: NOT NEEDED - pass local path directly to publish_content SUPPORTED FORMATS: • Images: jpg, png, gif, webp (max 10MB) • Videos: mp4, mov, webm (max 100MB) Returns { url: 'https://...' } for use in publish_content mediaUrl parameter.

  Connector

• scaffold_pipeline_workflow

  Keploy

  Generate the exact CI workflow YAML to add keploy sandbox tests to a pull-request pipeline, and tell you where to write it. Use this when the dev asks to "add keploy sandbox tests to my pipeline" / "wire keploy into CI" / "run keploy on PR" / "add a CI job for keploy" — the server emits the file contents verbatim so you don't have to compose the flag list yourself. ===== GOAL ===== Write a CI workflow file that runs `keploy test sandbox --cloud-app-id <uuid> --app-url <url>` on pull requests and gates the PR on the result. NEVER kick off an actual test run in this flow — it is pure file authoring, ends with the file on disk. DO NOT fire replay_sandbox_test, record_sandbox_test, replay_test_suite, or any other run-starting MCP tool here. ===== HOW (absolute) ===== Call this tool. It returns { file_path, content, summary }. Write the "content" to "file_path" VERBATIM via your Write tool — NO flag renames, NO flag removals, NO step reordering, NO synthesis. The server owns the YAML template; your job is only to (1) resolve the inputs from the repo and api-server and (2) Write the returned content. Do NOT compose the YAML yourself from general knowledge — flag drift (missing --cloud-app-id, inventing --app) is the most common bug when Claude improvises. DO NOT ASK the dev for confirmation before writing. Resolve everything from the repo + api-server, pick the GitHub Actions default, call this tool, Write the file. The dev's prompt is already the go-ahead. ===== STEPS ===== 1. DETECT THE CI SYSTEM: * Default = GitHub Actions (biggest share). File = .github/workflows/keploy-sandbox.yml. * If .gitlab-ci.yml exists → GitLab (not yet supported by this tool; tell the dev and stop). * If .circleci/config.yml exists → Circle (not yet supported; tell the dev and stop). * Otherwise → GitHub Actions. 2. RESOLVE VALUES by calling MCP tools + reading the repo: * app_id: call listApps({q: "<cwd basename>"}). Exactly one → use its id. Multiple → pick the one whose name most specifically matches the repo's primary service (e.g. "orderflow.producer" wins over "orderflow" when there's a ./producer directory); mention which you picked in the final message. Zero → stop and tell the dev to create the app + rerecord first. * suite_ids: DO NOT pass this arg by default. An empty suite_ids means the CLI resolves "every linked sandbox suite for the app" at CI run time — which is what you want (new suites auto-pick up without workflow edits). The tool still verifies there's ≥1 linked suite at scaffold time so the first PR run doesn't fail empty-handed. Only pass suite_ids when the dev explicitly narrows ("run only the auth suite in CI"); don't pin "all current suites" — that's staleness waiting to happen. * compose_file: READ THE REPO. Default is docker-compose.yml. AVOID passing a docker-compose-keploy.yaml variant that has `networks: default: external: true` — those variants only work locally, where another compose run has already created the external network. In CI the runner starts clean and `external: true` fails with "network not found". If the primary docker-compose.yml brings up the full app (deps + app service), use it end-to-end. * app_service, container_name, app_port: read from the SAME compose_file you picked above. app_service = the service key (e.g. "producer"); container_name = that service's container_name: field in that same compose file (e.g. "orderflow-producer" if compose_file=docker-compose.yml, but "producer" if compose_file=docker-compose-keploy.yaml — THESE DIFFER, pick consistently); app_port = the host-side of its ports: mapping. * app_url = http://localhost:<app_port>. The tool derives this; you don't pass it separately. 3. CALL THIS TOOL with app_id, app_service, container_name, app_port, compose_file (and suite_ids only if the dev explicitly narrowed scope). It returns { file_path, content, summary }. Write the "content" to the "file_path" VERBATIM. ===== FLAG NAME RULES (absolute, do not drift when reviewing the output) ===== * `--cloud-app-id` ← NOT `--app-id`. The OSS config has an `appId` uint64 field that viper maps `--app-id` into; passing a UUID there fails with "invalid syntax" before RunE runs. * `keploy test sandbox --cloud-app-id <uuid> --app-url <url>` ← the CI form. NOT `keploy test --cloud-app-id` (must be `test sandbox` — the headless flags live on the sandbox subcommand only), NOT `keploy test-suite run` (that command doesn't exist). There is NO `--pipeline` flag. * Install URL = `https://keploy.io/ent/install.sh` ← NOT `https://keploy.io/install.sh` (OSS; no sandbox subcommand at all), NOT a github.com/keploy/keploy release tarball. If the server-emitted content ever disagrees with these rules, trust the server output and file a bug — don't edit the YAML. ===== RESOLUTION ARGS ===== * Pass either app_id (explicit UUID) or app_name_hint (substring; server does listApps and requires exactly one match). * Pass app_service (docker-compose service name), container_name (from compose container_name: field read from the SAME compose_file arg), and app_port (HTTP port the service exposes). * compose_file is optional, defaults to "docker-compose.yml". If the repo has a -keploy.yaml variant with `external: true` networks, do NOT point compose_file at it — it won't work in CI. * suite_ids is optional and should be LEFT BLANK by default — the CLI resolves every linked suite at run time. Only pin an explicit list when the dev narrows scope. ===== FINAL RESPONSE — three short sections, no questions ===== ### Created | File | Lines | | --- | --- | | .github/workflows/keploy-sandbox.yml | N | ### Summary - App: <name> (<app_id>), <N> linked suites replayed on every PR - Trigger: pull_request → main, + manual workflow_dispatch - Failure on any suite gates the PR (non-zero exit from the CLI) ### Before the first run, add this GitHub secret - `KEPLOY_API_KEY` — at https://github.com/<owner>/<repo>/settings/secrets/actions/new (self-hosted users — point at your own api-server by building the enterprise binary with -X main.api_server_uri=<url>; there is no runtime env override on the released binary.) This tool does NOT run anything. It only generates file contents.

  Connector

• update_user

  mcp

  Update a database user for a Cloud SQL instance. A common use case for the `update_user` is to grant a user the `cloudsqlsuperuser` role, which can provide a user with many required permissions. This tool only supports updating users to assign database roles. * This tool returns a long-running operation. Use the `get_operation` tool to poll its status until the operation completes. * Before calling the `update_user` tool, always check the existing configuration of the user such as the user type with `list_users` tool. * As a special case for MySQL, if the `list_users` tool returns a full email address for the `iamEmail` field, for example `{name=test-account, iamEmail=test-account@project-id.iam.gserviceaccount.com}`, then in your `update_user` request, use the full email address in the `iamEmail` field in the `name` field of your toolrequest. For example, `name=test-account@project-id.iam.gserviceaccount.com`. Key parameters for updating user roles: * `database_roles`: A list of database roles to be assigned to the user. * `revokeExistingRoles`: A boolean field (default: false) that controls how existing roles are handled. How role updates work: 1. **If `revokeExistingRoles` is true:** * Any existing roles granted to the user but NOT in the provided `database_roles` list will be REVOKED. * Revoking only applies to non-system roles. System roles like `cloudsqliamuser` etc won't be revoked. * Any roles in the `database_roles` list that the user does NOT already have will be GRANTED. * If `database_roles` is empty, then ALL existing non-system roles are revoked. 2. **If `revokeExistingRoles` is false (default):** * Any roles in the `database_roles` list that the user does NOT already have will be GRANTED. * Existing roles NOT in the `database_roles` list are KEPT. * If `database_roles` is empty, then there is no change to the user's roles. Examples: * Existing Roles: `[roleA, roleB]` * Request: `database_roles: [roleB, roleC], revokeExistingRoles: true` * Result: Revokes `roleA`, Grants `roleC`. User roles become `[roleB, roleC]`. * Request: `database_roles: [roleB, roleC], revokeExistingRoles: false` * Result: Grants `roleC`. User roles become `[roleA, roleB, roleC]`. * Request: `database_roles: [], revokeExistingRoles: true` * Result: Revokes `roleA`, Revokes `roleB`. User roles become `[]`. * Request: `database_roles: [], revokeExistingRoles: false` * Result: No change. User roles remain `[roleA, roleB]`.

  Connector

• tfdeploy

  InsideOut (Riley)

  WORKFLOW: Step 4 of 4 - Deploy infrastructure to the cloud Deploy infrastructure by starting a Terraform job for an InsideOut session. This tool initiates the actual deployment process after Terraform files have been generated. IMPORTANT: This starts a long-running job (15+ minutes). Use tfstatus to monitor progress. SINGLE-FLIGHT: only one TF job (apply/plan/destroy/drift) runs per session at a time. If another job is already in flight, tfdeploy returns tf_job_conflict with the live job_id — attach with tfstatus/tflogs instead of retrying, or pass force_new=true to override. Returns confirmation that the deployment has started. REQUIRES: session_id from convoopen response (format: sess_v2_...). OPTIONAL: plan_id (string) — Apply a previously created plan from tfplan. Preview-then-apply workflow: tfplan → tflogs (review) → tfdeploy(plan_id=...). OPTIONAL: sandbox (boolean, default false) — deploys real generated Terraform. Set to true for cheap sandbox template (testing only). OPTIONAL: ignore_drift (boolean, default false) - when true, proceeds with deploy even if infrastructure drift is detected. By default, deploys fail on drift. Use after reviewing drift details via tfdrift or tflogs. OPTIONAL: force_new (boolean, default false) - bypass the session-level single-flight guard. Use only when the existing run is provably wedged. CREDENTIAL FLOW (if credentials are missing): 1. Response includes a connect_url — present it to the user 2. Call credawait(session_id=...) to poll for credentials 3. When credawait returns success, retry tfdeploy Do NOT call credawait without first showing the connect URL to the user.

  Connector

• nausika_seasonal_forecast

  Nausika

  Seasonal outlook at (latitude, longitude in decimal degrees), ECMWF SEAS5 (1–7 month) via Open-Meteo. Returns ENSEMBLE-MEAN absolutes as DAILY + MONTHLY aggregates. Core: wind, SST, temperature, precipitation, cloud cover; add more via variables[]. NO waves — use nausika_marine_forecast for short-term + waves. Example: latitude=40.12, longitude=9.01, start_date="2026-08-01", end_date="2026-08-31". Chain nausika_geocode for place→coordinates. Constraints: end_date ≤ today+210d, range ≤ 215d.

  Connector

• architect.certify

  AI Design Blueprint

  Pro/Teams — second-pass adversarial certification of an architect.validate run that scored production_ready (A or B first-pass tier). Mints the certified production_ready badge when both reviewers sign off; caps the run to C/emerging when the second pass surfaces a missed production_blocker. MANDATORY DOCTRINE RULE (load-bearing): the badge certifies the EXACT code that produced the validate run_id, NOT 'this codebase' in general. If you modify, fix, or iterate the code between architect.validate and architect.certify — even a single character — cert rejects with code_fingerprint_mismatch. Fixing the code voids the run. The recovery path is always: edit code → architect.validate → fresh run_id → architect.certify on the fresh run. Do NOT cert from a stale run_id after iteration; ask the user to re-validate first. WHEN TO CALL: only after architect.validate returned tier=production_ready AND the user wants the certified badge AND the code has not been touched since the validate run. NOT for tier=draft/emerging/not_applicable runs (typed rejections fire — see below). NOT idempotent across attempts: each call is one of the 3 attempts in the retry budget. BEHAVIOR: atomic one-shot single LLM call, ~60-180s server-side at high reasoning effort (small payloads finish faster; observed p99 ~250s; server-side budget is 20 min, ~5× observed max). Exceeds typical MCP-client tool-call idle budget (~60s in Claude Code), so the FIRST notifications/progress event fires at t=0 carrying the run_id. The run is atomic by contract — no in_progress lifecycle, no cancellation, no resume. Updates the persisted run's result_json (public review URL + me.validation_history(run_id=...) reflect the cert outcome). ELIGIBILITY GATE (typed rejection enum on failure): caller must own the run, tier=production_ready, less than 24h old, not already certified, within cert retry budget (max 3 attempts), no other cert call in flight for the same run_id, code fingerprint must match the validated code, AND the submitted payload must be cert-payload-complete (see Payload Completeness below — cert rejects pre-LLM with `payload_incomplete` when an imported module's surface isn't visible in the validate payload that produced this run_id). Rejection reasons (typed Literal): auth_required, paid_plan_required, run_not_found, not_run_owner, not_eligible_tier, not_agentic_component (tier=not_applicable runs), already_certified, certification_age_exceeded, retry_budget_exhausted, code_fingerprint_mismatch, code_fingerprint_missing, code_not_on_file (caller omitted `code` argument AND the 24h cert-retry hold for this run has expired or was never written. Recovery: re-run architect.certify from the same MCP session that ran architect.validate, passing the code explicitly — the server never persists code by design), payload_incomplete (submitted/validated payload imports modules whose contents aren't visible — cert refuses pre-LLM to prevent a false-precision downgrade. Recovery: re-validate with verbatim public-surface stubs for every imported module, then re-cert on the fresh run_id. Empirically validated: PR #157 iter8/iter9 cert rejections were exactly this class — code on disk was correct, the submitted payload merely omitted module visibility), cert_consensus_score_below_threshold (consensus_median<75 — consensus runs only), cert_consensus_unstable_blocker (any principle mode_stability<80% — consensus runs only), run_state_corrupt, cert_persistence_failed, cert_in_flight (a prior architect.certify call on this run_id is still running. Poll me.validation_history for the verdict; do not retry until it resolves). PAYLOAD COMPLETENESS (load-bearing for cert eligibility): the cert reviewer reads the EXACT payload that produced the validate run_id. Imported modules whose surface isn't present in the payload cause pre-LLM `payload_incomplete` refusal. Avoidance — when validating with intent to cert, bundle public-surface stubs for every imported module: `from sqlalchemy.exc import SQLAlchemyError` → include a stub class; `from app.db import models` → include a `class models:` namespace stub with the columns/methods you reference; module-level imports of `dataclass`, `Literal`, `json`, `datetime`, `timezone` MUST also be in the payload (cert correctly catches when they're omitted — code would NameError on import). 'Submit Like Production': the payload should be the code as it would actually run, not a compressed sketch. PRE-LLM REJECTION AUDIT TRAIL: when cert rejects before the LLM call (payload_incomplete, code_fingerprint_mismatch, etc.), `certification_attempts=[]` on the response — no attempt landed in the retry budget, no LLM hop occurred. The rejection envelope's `rejection_reason` + `guidance` are the actionable surface. (Audit-trail UI surfacing of pre-LLM rejections is tracked in the platform self-audit set as anomaly #5; out of scope for the cert tool itself.) INPUTS: re-send the SAME code that produced the run_id (the architect persists findings + recommendations, never code, by design — privacy-preserving). Server compares the submitted code's SHA-256 fingerprint to the stored fingerprint and rejects mismatches. Auth: Bearer <token>, Pro or Teams plan required. UK/EU data residency (Cloud Run europe-west2). Code processed transiently by OpenAI (no-training-on-API-data) and dropped; payloads JSON-escaped + delimited as inert untrusted data — prompt-injection inside code is ignored. RECOVERY: if your MCP client closes the tool-call early, recover the cert verdict via me.validation_history(run_id=<that-id>) once the server-side LLM call lands — same Bearer token, same pattern as architect.validate. If the cert call fails outright (provider error, persistence error), a fresh architect.certify is the recovery path; the eligibility gate enforces the 3-attempt retry budget. For long-running cert workflows the answer is to re-validate, not to make this tool stateful. OUTCOMES: certification_status ∈ {confirmed_production_ready (badge mints), downgraded_to_emerging (cert review surfaced a missed production_blocker, tier capped at C/emerging), unavailable_provider_error (LLM call failed, retry within budget)}. Cert findings + summary + attempt history surfaced on the persisted run for full inspectability.

  Connector

• tfdeploy

  InsideOut (Riley)

  WORKFLOW: Step 4 of 4 - Deploy infrastructure to the cloud Deploy infrastructure by starting a Terraform job for an InsideOut session. This tool initiates the actual deployment process after Terraform files have been generated. IMPORTANT: This starts a long-running job (15+ minutes). Use tfstatus to monitor progress. SINGLE-FLIGHT: only one TF job (apply/plan/destroy/drift) runs per session at a time. If another job is already in flight, tfdeploy returns tf_job_conflict with the live job_id — attach with tfstatus/tflogs instead of retrying, or pass force_new=true to override. Returns confirmation that the deployment has started. REQUIRES: session_id from convoopen response (format: sess_v2_...). OPTIONAL: plan_id (string) — Apply a previously created plan from tfplan. Preview-then-apply workflow: tfplan → tflogs (review) → tfdeploy(plan_id=...). OPTIONAL: sandbox (boolean, default false) — deploys real generated Terraform. Set to true for cheap sandbox template (testing only). OPTIONAL: ignore_drift (boolean, default false) - when true, proceeds with deploy even if infrastructure drift is detected. By default, deploys fail on drift. Use after reviewing drift details via tfdrift or tflogs. OPTIONAL: force_new (boolean, default false) - bypass the session-level single-flight guard. Use only when the existing run is provably wedged. CREDENTIAL FLOW (if credentials are missing): 1. Response includes a connect_url — present it to the user 2. Call credawait(session_id=...) to poll for credentials 3. When credawait returns success, retry tfdeploy Do NOT call credawait without first showing the connect URL to the user.

  Connector

• architect.certify

  AI Design Blueprint Doctrine

  Pro/Teams — second-pass adversarial certification of an architect.validate run that scored production_ready (A or B first-pass tier). Mints the certified production_ready badge when both reviewers sign off; caps the run to C/emerging when the second pass surfaces a missed production_blocker. MANDATORY DOCTRINE RULE (load-bearing): the badge certifies the EXACT code that produced the validate run_id, NOT 'this codebase' in general. If you modify, fix, or iterate the code between architect.validate and architect.certify — even a single character — cert rejects with code_fingerprint_mismatch. Fixing the code voids the run. The recovery path is always: edit code → architect.validate → fresh run_id → architect.certify on the fresh run. Do NOT cert from a stale run_id after iteration; ask the user to re-validate first. WHEN TO CALL: only after architect.validate returned tier=production_ready AND the user wants the certified badge AND the code has not been touched since the validate run. NOT for tier=draft/emerging/not_applicable runs (typed rejections fire — see below). NOT idempotent across attempts: each call is one of the 3 attempts in the retry budget. BEHAVIOR: atomic one-shot single LLM call, ~60-180s server-side at high reasoning effort (small payloads finish faster; observed p99 ~250s; server-side budget is 20 min, ~5× observed max). Exceeds typical MCP-client tool-call idle budget (~60s in Claude Code), so the FIRST notifications/progress event fires at t=0 carrying the run_id. The run is atomic by contract — no in_progress lifecycle, no cancellation, no resume. Updates the persisted run's result_json (public review URL + me.validation_history(run_id=...) reflect the cert outcome). ELIGIBILITY GATE (typed rejection enum on failure): caller must own the run, tier=production_ready, less than 24h old, not already certified, within cert retry budget (max 3 attempts), no other cert call in flight for the same run_id, code fingerprint must match the validated code, AND the submitted payload must be cert-payload-complete (see Payload Completeness below — cert rejects pre-LLM with `payload_incomplete` when an imported module's surface isn't visible in the validate payload that produced this run_id). Rejection reasons (typed Literal): auth_required, paid_plan_required, run_not_found, not_run_owner, not_eligible_tier, not_agentic_component (tier=not_applicable runs), already_certified, certification_age_exceeded, retry_budget_exhausted, code_fingerprint_mismatch, code_fingerprint_missing, code_not_on_file (caller omitted `code` argument AND the 24h cert-retry hold for this run has expired or was never written. Recovery: re-run architect.certify from the same MCP session that ran architect.validate, passing the code explicitly — the server never persists code by design), payload_incomplete (submitted/validated payload imports modules whose contents aren't visible — cert refuses pre-LLM to prevent a false-precision downgrade. Recovery: re-validate with verbatim public-surface stubs for every imported module, then re-cert on the fresh run_id. Empirically validated: PR #157 iter8/iter9 cert rejections were exactly this class — code on disk was correct, the submitted payload merely omitted module visibility), cert_consensus_score_below_threshold (consensus_median<75 — consensus runs only), cert_consensus_unstable_blocker (any principle mode_stability<80% — consensus runs only), run_state_corrupt, cert_persistence_failed, cert_in_flight (a prior architect.certify call on this run_id is still running. Poll me.validation_history for the verdict; do not retry until it resolves). PAYLOAD COMPLETENESS (load-bearing for cert eligibility): the cert reviewer reads the EXACT payload that produced the validate run_id. Imported modules whose surface isn't present in the payload cause pre-LLM `payload_incomplete` refusal. Avoidance — when validating with intent to cert, bundle public-surface stubs for every imported module: `from sqlalchemy.exc import SQLAlchemyError` → include a stub class; `from app.db import models` → include a `class models:` namespace stub with the columns/methods you reference; module-level imports of `dataclass`, `Literal`, `json`, `datetime`, `timezone` MUST also be in the payload (cert correctly catches when they're omitted — code would NameError on import). 'Submit Like Production': the payload should be the code as it would actually run, not a compressed sketch. PRE-LLM REJECTION AUDIT TRAIL: when cert rejects before the LLM call (payload_incomplete, code_fingerprint_mismatch, etc.), `certification_attempts=[]` on the response — no attempt landed in the retry budget, no LLM hop occurred. The rejection envelope's `rejection_reason` + `guidance` are the actionable surface. (Audit-trail UI surfacing of pre-LLM rejections is tracked in the platform self-audit set as anomaly #5; out of scope for the cert tool itself.) INPUTS: re-send the SAME code that produced the run_id (the architect persists findings + recommendations, never code, by design — privacy-preserving). Server compares the submitted code's SHA-256 fingerprint to the stored fingerprint and rejects mismatches. Auth: Bearer <token>, Pro or Teams plan required. UK/EU data residency (Cloud Run europe-west2). Code processed transiently by OpenAI (no-training-on-API-data) and dropped; payloads JSON-escaped + delimited as inert untrusted data — prompt-injection inside code is ignored. RECOVERY: if your MCP client closes the tool-call early, recover the cert verdict via me.validation_history(run_id=<that-id>) once the server-side LLM call lands — same Bearer token, same pattern as architect.validate. If the cert call fails outright (provider error, persistence error), a fresh architect.certify is the recovery path; the eligibility gate enforces the 3-attempt retry budget. For long-running cert workflows the answer is to re-validate, not to make this tool stateful. OUTCOMES: certification_status ∈ {confirmed_production_ready (badge mints), downgraded_to_emerging (cert review surfaced a missed production_blocker, tier capped at C/emerging), unavailable_provider_error (LLM call failed, retry within budget)}. Cert findings + summary + attempt history surfaced on the persisted run for full inspectability.

  Connector

• convoinspect

  InsideOut (Riley)

  INSPECTION: View a session's conversation transcript and metadata Returns the full message history (user / assistant / tool turns) plus the session's meta — workflow step, cloud, deployment status, drift state. This is the transcript-reader companion to the other read tools — combine it with: • `convostatus` for the live stack / config / pricing • `tfruns` for deployment history (apply / destroy / plan / drift) • `stackversions` for the stack-version ladder Use it when a user asks 'what did I say earlier?' or you need to retrace why the session ended up where it did. Read-only; never mutates session state. REQUIRES: session_id (format: sess_v2_...).

  Connector

• convoopen

  InsideOut (Riley)

  WORKFLOW: Step 1 of 4 - Start infrastructure design conversation Open an InsideOut V2 session and receive the assistant's intro message. The response contains a clean message from Riley (the infrastructure advisor) - display it to the user. ⚠️ Riley will ask questions - forward these to the user, DO NOT answer on their behalf. CRITICAL: This tool returns a session_id in the response metadata. You MUST use this session_id for ALL subsequent tool calls (convoreply, tfgenerate, tfdeploy, etc.). ⚠️ The session_id includes a ?token=... suffix (format: sess_v2_xxx?token=yyy) which is part of the session credential — without it, downstream tools fall back to a tokenless connect URL that 401s. Always pass session_id verbatim to subsequent tools and to the user; do NOT shorten, paraphrase, or strip the ?token= portion when summarizing the session in chat or in your own scratch notes. Use when the user mentions keywords like: 'setup my cloud infra', 'provision infrastructure', 'deploy infra', 'start insideout', 'use insideout', or similar intent to begin infra setup. OPTIONAL: project_context (string) - General tech stack summary so Riley can skip discovery questions and jump to recommendations. The agent should confirm this with the user before sending. Include whichever apply: language/framework, databases/services, container usage, existing IaC, CI/CD platform, cloud provider, Kubernetes usage, what the project does. Example: 'Next.js 14 + TypeScript, PostgreSQL, Redis, Docker Compose, deployed to AWS ECS, GitHub Actions CI/CD, ~50k MAU'. NEVER include credentials, secrets, API keys, PII, source code, or internal URLs/IPs -- only general metadata summaries useful to a cloud architect agent. IMPORTANT: source (string) - You MUST set this to identify which IDE/tool you are. Auto-detect from your environment: 'claude-code', 'codex', 'antigravity', 'kiro', 'vscode', 'web', 'mcp'. If unsure, use the name of your IDE/tool in lowercase. Do NOT omit this — it controls the 'Open {IDE}' button on the credential connect screen. OPTIONAL: github_username (string) - GitHub username for deploy commit attribution. Pre-populates the GitHub username field on the connect page. 💡 TIP: Examine workflow.usage prompt for more context on how to properly use these tools.

  Connector

• convoopen

  InsideOut (Riley)

  WORKFLOW: Step 1 of 4 - Start infrastructure design conversation Open an InsideOut V2 session and receive the assistant's intro message. The response contains a clean message from Riley (the infrastructure advisor) - display it to the user. ⚠️ Riley will ask questions - forward these to the user, DO NOT answer on their behalf. CRITICAL: This tool returns a session_id in the response metadata. You MUST use this session_id for ALL subsequent tool calls (convoreply, tfgenerate, tfdeploy, etc.). ⚠️ The session_id includes a ?token=... suffix (format: sess_v2_xxx?token=yyy) which is part of the session credential — without it, downstream tools fall back to a tokenless connect URL that 401s. Always pass session_id verbatim to subsequent tools and to the user; do NOT shorten, paraphrase, or strip the ?token= portion when summarizing the session in chat or in your own scratch notes. Use when the user mentions keywords like: 'setup my cloud infra', 'provision infrastructure', 'deploy infra', 'start insideout', 'use insideout', or similar intent to begin infra setup. OPTIONAL: project_context (string) - General tech stack summary so Riley can skip discovery questions and jump to recommendations. The agent should confirm this with the user before sending. Include whichever apply: language/framework, databases/services, container usage, existing IaC, CI/CD platform, cloud provider, Kubernetes usage, what the project does. Example: 'Next.js 14 + TypeScript, PostgreSQL, Redis, Docker Compose, deployed to AWS ECS, GitHub Actions CI/CD, ~50k MAU'. NEVER include credentials, secrets, API keys, PII, source code, or internal URLs/IPs -- only general metadata summaries useful to a cloud architect agent. IMPORTANT: source (string) - You MUST set this to identify which IDE/tool you are. Auto-detect from your environment: 'claude-code', 'codex', 'antigravity', 'kiro', 'vscode', 'web', 'mcp'. If unsure, use the name of your IDE/tool in lowercase. Do NOT omit this — it controls the 'Open {IDE}' button on the credential connect screen. OPTIONAL: github_username (string) - GitHub username for deploy commit attribution. Pre-populates the GitHub username field on the connect page. 💡 TIP: Examine workflow.usage prompt for more context on how to properly use these tools.

  Connector

• gcpinspect

  InsideOut (Riley)

  INSPECTION: Inspect GCP infrastructure for a deployed project ⚠️ **PREREQUISITE**: This tool requires a prior deployment ATTEMPT (successful or failed). Check convostatus for hasDeployAttempt=true before calling. Works even after failed deploys to inspect orphaned resources. Inspect deployed GCP resources after a deployment attempt. Use this tool when the user asks about the status or details of their deployed GCP infrastructure. It fetches temporary read-only credentials securely and queries the GCP API directly. RESPONSE TIERS (default is summary for token efficiency): - Summary (default): Key fields only (~500 tokens). Set detail=false, raw=false or omit both. - Detail: Full metadata for a specific resource. Set detail=true + resource filter. - Raw: Complete unprocessed API response. Set raw=true. REQUIRES: session_id from convoopen response (format: sess_v2_...). Supported services: apigateway, bastion, billing, cloudarmor, cloudbuild, cloudcdn, cloudfunctions, cloudkms, cloudlogging, cloudmonitoring, cloudrun, cloudsql, compute, firestore, gcs, gke, identityplatform, loadbalancer, memorystore, pubsub, secretmanager, vertexai, vpc For a specific service's actions, call with action="list-actions". METRICS: Use list-metrics to see available Cloud Monitoring metrics for any service (no credentials needed — progressive disclosure). Use get-metrics to retrieve time-series data. Optional filters JSON: {"hours":6,"period":300}. Label breakdowns: Cloud Functions (by status), Load Balancer/API Gateway (by response_code_class), Cloud CDN (by cache_result). Secret Manager get-metrics returns operational health (version count, replication, create time) — no time-series. Bastion is an alias for Compute Engine metrics (SSH connection count not available as a GCP metric). BILLING: Use service=billing to inspect GCP billing. Actions: get-billing-info (check if billing enabled, which billing account), get-budgets (list budget alerts for the project — auto-fetches billing account). Requires roles/billing.viewer IAM role. Required IAM roles: Monitoring Viewer (roles/monitoring.viewer) for metrics, Secret Manager Viewer (roles/secretmanager.viewer) for secret health, Billing Viewer (roles/billing.viewer) for billing. EXAMPLES: - gcpinspect(session_id=..., service="compute", action="list-instances") - gcpinspect(session_id=..., service="gke", action="list-clusters") - gcpinspect(session_id=..., service="cloudsql", action="get-metrics", filters="{\"hours\":6}") - gcpinspect(session_id=..., service="billing", action="get-billing-info")

  Connector

• import_data

  mcp

  Import data into a Cloud SQL instance. If the file doesn't start with `gs://`, then the assumption is that the file is stored locally. If the file is local, then the file must be uploaded to Cloud Storage before you can make the actual `import_data` call. To upload the file to Cloud Storage, you can use the `gcloud` or `gsutil` commands. Before you upload the file to Cloud Storage, consider whether you want to use an existing bucket or create a new bucket in the provided project. After the file is uploaded to Cloud Storage, the instance service account must have sufficient permissions to read the uploaded file from the Cloud Storage bucket. This can be accomplished as follows: 1. Use the `get_instance` tool to get the email address of the instance service account. From the output of the tool, get the value of the `serviceAccountEmailAddress` field. 2. Grant the instance service account the `storage.objectAdmin` role on the provided Cloud Storage bucket. Use a command like `gcloud storage buckets add-iam-policy-binding` or a request to the Cloud Storage API. It can take from two to up to seven minutes or more for the role to be granted and the permissions to be propagated to the service account in Cloud Storage. If you encounter a permissions error after updatingthe IAM policy, then wait a few minutes and try again. After permissions are granted, you can import the data. We recommend that you leave optional parameters empty and use the system defaults. The file type can typically be determined by the file extension. For example, if the file is a SQL file, `.sql` or `.csv` for CSV file. The following is a sample SQL `importContext` for MySQL. ``` { "uri": "gs://sample-gcs-bucket/sample-file.sql", "kind": "sql#importContext", "fileType": "SQL" } ``` There is no `database` parameter present for MySQL since the database name is expected to be present in the SQL file. Specify only one URI. No other fields are required outside of `importContext`. For PostgreSQL, the `database` field is required. The following is a sample PostgreSQL `importContext` with the `database` field specified. ``` { "uri": "gs://sample-gcs-bucket/sample-file.sql", "kind": "sql#importContext", "fileType": "SQL", "database": "sample-db" } ``` The `import_data` tool returns a long-running operation. Use the `get_operation` tool to poll its status until the operation completes.

  Connector

• gcpinspect

  InsideOut (Riley)

  INSPECTION: Inspect GCP infrastructure for a deployed project ⚠️ **PREREQUISITE**: This tool requires a prior deployment ATTEMPT (successful or failed). Check convostatus for hasDeployAttempt=true before calling. Works even after failed deploys to inspect orphaned resources. Inspect deployed GCP resources after a deployment attempt. Use this tool when the user asks about the status or details of their deployed GCP infrastructure. It fetches temporary read-only credentials securely and queries the GCP API directly. RESPONSE TIERS (default is summary for token efficiency): - Summary (default): Key fields only (~500 tokens). Set detail=false, raw=false or omit both. - Detail: Full metadata for a specific resource. Set detail=true + resource filter. - Raw: Complete unprocessed API response. Set raw=true. REQUIRES: session_id from convoopen response (format: sess_v2_...). Supported services: apigateway, bastion, billing, cloudarmor, cloudbuild, cloudcdn, cloudfunctions, cloudkms, cloudlogging, cloudmonitoring, cloudrun, cloudsql, compute, firestore, gcs, gke, identityplatform, loadbalancer, memorystore, pubsub, secretmanager, vertexai, vpc For a specific service's actions, call with action="list-actions". METRICS: Use list-metrics to see available Cloud Monitoring metrics for any service (no credentials needed — progressive disclosure). Use get-metrics to retrieve time-series data. Optional filters JSON: {"hours":6,"period":300}. Label breakdowns: Cloud Functions (by status), Load Balancer/API Gateway (by response_code_class), Cloud CDN (by cache_result). Secret Manager get-metrics returns operational health (version count, replication, create time) — no time-series. Bastion is an alias for Compute Engine metrics (SSH connection count not available as a GCP metric). BILLING: Use service=billing to inspect GCP billing. Actions: get-billing-info (check if billing enabled, which billing account), get-budgets (list budget alerts for the project — auto-fetches billing account). Requires roles/billing.viewer IAM role. Required IAM roles: Monitoring Viewer (roles/monitoring.viewer) for metrics, Secret Manager Viewer (roles/secretmanager.viewer) for secret health, Billing Viewer (roles/billing.viewer) for billing. EXAMPLES: - gcpinspect(session_id=..., service="compute", action="list-instances") - gcpinspect(session_id=..., service="gke", action="list-clusters") - gcpinspect(session_id=..., service="cloudsql", action="get-metrics", filters="{\"hours\":6}") - gcpinspect(session_id=..., service="billing", action="get-billing-info")

  Connector

• install_app

  BorealHost

  Install an app template on a VPS/Cloud site. Starts a background installation. Poll get_app_status() for progress. Requires: API key with write scope. VPS or Cloud plan only. Args: slug: Site identifier template: App template slug. Available: django, laravel, nextjs, nodejs, nuxtjs, rails, static, forge app_name: Short name for the app (2-50 chars, lowercase alphanumeric + hyphens). Used as subdomain: {app_name}.{site_domain} db_type: Database type. "none", "mysql", or "postgresql" (depends on template) domain: Custom domain override (default: {app_name}.{site_domain}) display_name: Human-friendly name (default: derived from app_name) Returns: {"id": "uuid", "app_name": "forge", "status": "installing", "message": "Installation started. Poll for progress."} Errors: FORBIDDEN: Plan does not support apps (shared plans) VALIDATION_ERROR: Invalid template, app_name, or duplicate name

  Connector

• upload_recording

  Keploy

  Atomic test set + cases + mocks + mappings ingest. Creates the test set row, every test case, every mock, and the mapping doc in one call. PREFER THE CLI FOR ON-DISK RECORDINGS. When the dev has a recorded test-set on disk (e.g. `./keploy/test-set-0/` produced by `keploy record`), invoke this via Bash instead — it streams bytes from disk to server in one HTTP round-trip: ``` keploy upload test-set \ --app <namespace.deployment> # or --cloud-app-id <uuid> --branch <uuid|name> # optional, find-or-create on name --test-set <path|name> # e.g. keploy/test-set-0 [--name <override>] # rename on the server ``` The CLI path runs in ~3 seconds for a typical recording; calling this MCP tool directly with the same bundle inlined as args takes minutes because Claude has to serialize ~10K+ tokens of YAML/JSON through tool_use. Reserve this MCP tool for cases where the data is already in conversation context (e.g. you just generated test cases programmatically and don't want to round-trip to disk). Each step is its own DB write; partial failure leaves earlier rows in place — callers can replay safely. `branch_id` is REQUIRED — direct writes to main via MCP are blocked. Every row lands on the branch overlay until merge. `test_cases[].mock_names` lists the mocks each case consumes; the server folds these into the mapping doc on upload. Returns { test_set, test_case_ids, mock_ids }.

  Connector

• credawait

  InsideOut (Riley)

  Wait for the user to securely connect their cloud account and subscribe to Luther Systems. Polls until credentials appear on the session. 🎯 USE THIS TOOL WHEN: tfdeploy returns an 'auth_required', 'no_credentials', or 'credentials_expired' error. The user needs to visit the connect URL to: 1. Connect their cloud credentials (AWS or GCP) 2. Sign up and subscribe to a Luther Systems plan (required for deployment) This secure connection allows InsideOut to deploy and manage infrastructure in the user's cloud account on their behalf. Credentials are handled securely and only used for deployment and management sessions. WORKFLOW: 1. FIRST: Present the connect URL and explanation to the user (from the tfdeploy error response) 2. THEN: Call this tool to begin polling for credentials 3. The user opens the URL in their browser to subscribe and add credentials 4. When credentials are found, inform the user and call tfdeploy to deploy IMPORTANT: Do NOT call this tool without first showing the connect URL to the user. The user needs to see the URL to complete the process. REQUIRES: session_id from convoopen response (format: sess_v2_...). OPTIONAL: cloud ('aws' or 'gcp'), timeout (integer, seconds to wait, default 300, max 600).

  Connector