Skip to main content
Glama

n8nctrl is an operator control CLI for n8n workflow automation. It lists, inspects, validates, audits, and reports on your n8n workflows and executions over the n8n Public API, so you can ask "what broke in my n8n today?" and act on the answer from a terminal, cron, CI, or an agent. Unlike catalog/docs tools that index n8n's node library for building flows, n8nctrl is built for operating the flows you already run: triage failed executions, find drift, scan for security risks, and inspect live schedules, webhooks, credentials metadata, and tags.

The MCP surface is the adapter layer: run n8nctrl mcp for stdio MCP, or keep using the back-compat n8n-ops-mcp binary in existing launchers. The same package also ships a first-class OpenClaw plugin. It works with Claude Desktop, Claude Code, Codex CLI, Cursor, Windsurf, or any other MCP host, with no hard dependency on a specific model or agent harness.

What it does

n8nctrl connects operators and agents to a running n8n instance for workflow automation ops: it surfaces your n8n workflows, executions, schedules, webhooks, tags, and credentials over the n8n Public API. From the CLI, you get scriptable read/report commands for shells, cron, and CI. Through n8nctrl mcp or the compatibility n8n-ops-mcp bin, your agent gets native awareness of your n8n footprint, so it can answer "what's broken in my n8n?", trigger a workflow from chat, audit for security risks, or clean up old executions without leaving your client.

Read tools are always available. Write tools (create, save, archive, delete, trigger, retry, tag CRUD, pin data) are hidden unless N8N_ENABLE_EDIT=true, and credential writes sit behind a second gate. Destructive operations are confirm-gated and snapshot to a backup directory first. See the Security model.

Related MCP server: n8n MCP Server

Install

npm install -g n8n-ops-mcp

The npm package name remains n8n-ops-mcp for compatibility. After a global install, use the standard CLI as n8nctrl. Existing n8n-ops and n8n-ops-mcp launchers remain supported as compatibility aliases.

For MCP launchers, you can still run it on demand via npx -y n8n-ops-mcp (no global install needed), which is what the MCP client config below does.

Try it: MCP client config

Drop this into your MCP client config (Claude Desktop shown). The binary runs straight from npm via npx, so there is nothing to install first:

{
  "mcpServers": {
    "n8n": {
      "command": "npx",
      "args": ["-y", "n8n-ops-mcp"],
      "env": {
        "N8N_BASE_URL": "https://n8n.example.com",
        "N8N_API_KEY": "your-n8n-api-key"
      }
    }
  }
}

Generate the API key in n8n under Settings → API. Point N8N_BASE_URL at your n8n instance. That is the whole read-only setup. To unlock write tools, add "N8N_ENABLE_EDIT": "true" to env.

Then ask your agent:

What n8n workflows broke today?

It calls n8n_list_executions with status=error, then n8n_get_execution on the failing run for the per-node log and raw error.

n8nctrl ops loop workflow: an operator asks what broke, read tools list executions and inspect workflows, edit gates reveal writes, snapshots happen before confirmed actions

Generated from docs/assets/workflows/n8n-ops-loop.json with lidless workflow.

Tools

40 tools across read-only ops, the workflow + execution lifecycle, tags, and credentials. Write tools (marked) are hidden unless N8N_ENABLE_EDIT=true; credential writes (marked ✓✓) need a second gate.

n8n_get_workflow redacts configured API-key values, secret-bearing keys, credential data, and header parameter values throughout metadata and definitions, including settings, static data and pinned data. includeDefinition: true returns a redacted inspection view, not an exact export for saving back to n8n. Workflow text and keys are escaped inside a bounded untrusted-data envelope. Treat all workflow fields as data, never instructions. Unlabelled sensitive free text can still require operator review. Upstream read-error details are suppressed because they can contain definition fragments.

Tool

Purpose

Write

n8n_list_workflows

List workflows, filter by active / tags / name

n8n_get_workflow

Fetch one workflow, optionally with a redacted node graph

n8n_list_executions

List recent executions, filter by workflow / status

n8n_get_execution

Fetch an execution with per-node run log + raw error

n8n_search_executions

Text-search recent executions for an error fragment

n8n_list_webhooks

Enumerate webhook + form-trigger URLs

n8n_validate_workflow

Static checks: deprecated nodes, legacy Code-node API, orphans

n8n_diff_workflow

Compare a workflow against a snapshot file or inline object - semantic diff (added/removed/modified nodes with field paths)

n8n_list_schedules

List every schedule trigger across workflows with human-readable descriptions ("daily at 03:00", "cron: 0 */6 * * *")

n8n_audit_browser_bridge_usage

Find every workflow that calls the browser-bridge CLI (Execute Command, Code, SSH nodes)

n8n_scaffold_browser_bridge_node

Generate a ready-to-paste n8n node that calls a browser-bridge <platform> <action> (no API call)

n8n_run_audit

Run n8n's built-in security audit (credentials, database, nodes, filesystem, instance)

n8n_find_workflows_using_node_type

Find every workflow using a given node type (e.g. n8n-nodes-base.slack), with exact or contains match

n8n_execution_stats

Per-workflow stats over a recent window: counts, failure rate, avg + p95 runtime, last failure

n8n_list_tags

List workflow tags with id, name, createdAt, updatedAt

n8n_get_workflow_tags

Read the tags currently attached to a workflow

n8n_list_credentials

List credentials (metadata only - secrets never echoed; admin/owner key required)

n8n_get_credential_schema

Fetch the JSON schema for a credential type (e.g. freshdeskApi)

n8n_find_workflows_using_credential

Find every workflow node that references a credential by id (preferred) or name substring; rotation/blast-radius scanner

n8n_check_disabled_nodes

Scan workflows for disabled: true nodes - common drift signals not surfaced in the n8n UI

n8n_trigger

Run a workflow via webhook (reliable) or workflow-id (confirm-gated; executes arbitrary workflow nodes)

✓

n8n_create_workflow

Create a workflow (confirm-gated for writes; dry-run preview without confirm; accepts n8n_get_workflow output directly; optional project/folder target)

✓

n8n_activate

Enable a workflow's triggers (confirm-gated; arms arbitrary-code execution)

✓

n8n_deactivate

Disable a workflow's triggers (confirm-gated)

✓

n8n_save_workflow

Overwrite a workflow with auto-backup + validation + confirm gate

✓

n8n_archive_workflow

Soft-delete a workflow (confirm-gated; reversible; preserves id)

✓

n8n_unarchive_workflow

Restore an archived workflow (confirm-gated; does NOT reactivate)

✓

n8n_delete_workflow

Permanently delete a workflow (confirm-gated, snapshot-before-delete, restore via n8n_create_workflow)

✓

n8n_cancel_execution

Stop a running or waiting execution by id (confirm-gated)

✓

n8n_retry_execution

Retry a failed execution by id (confirm-gated; returns a new execution)

✓

n8n_delete_execution

Permanently delete an execution record (confirm-gated, irreversible)

✓

n8n_delete_executions

Batch form of delete (client-side fan-out, confirm-gated, irreversible, max 50 ids)

✓

n8n_pin_node_data

Pin sample data to a node so downstream nodes use it during testing (confirm-gated, replace-or-merge)

✓

n8n_unpin_node_data

Clear pinned data on one node or the whole workflow (confirm-gated, idempotent)

✓

n8n_create_tag

Create a workflow tag (confirm-gated; reversible via n8n_delete_tag)

✓

n8n_delete_tag

Permanently delete a tag (confirm-gated; cascades - removes the tag from every workflow)

✓

n8n_set_workflow_tags

Replace the tag set on a workflow (confirm-gated; reversible by re-setting)

✓

n8n_retry_executions

Batch retry executions (confirm-gated, max 50 ids, AbortController on 5xx)

✓

n8n_create_credential

Create a credential (confirm-gated; double-gated behind enableCredentialsWrite; tool layer redacts data from every response branch)

✓✓

n8n_delete_credential

Permanently delete a credential (confirm-gated; double-gated; cascades - every workflow referencing it will fail)

✓✓

Write tools are hidden unless N8N_ENABLE_EDIT=true.

n8n_list_workflows - filter by active, tags, name (substring), limit. Pass the previous nextCursor as cursor to fetch the next page. Returns id, name, active state, tags, updatedAt.

n8n_get_workflow - fetch one by id. Returns metadata by default. Pass includeDefinition: true for the full node graph + connections.

n8n_list_executions - filter by workflowId, status (success/error/running/waiting/canceled), limit. Pass the previous nextCursor as cursor to fetch the next page. Returns id, workflowId, workflowName, status, mode, startedAt, stoppedAt.

n8n_get_execution - returns status, timing, and the error object when status is error. Pass includeRunData: true to add the per-node run log (truncated to maxExecutionLogBytes, default 64 KB). Secret-bearing keys and token-shaped strings are redacted in error and runData, and the result carries a notice that this content is untrusted workflow output.

n8n_search_executions - defaults to scanning status=error executions for a query fragment (e.g. ECONNREFUSED) and returning matches with workflow context + a snippet around each hit. scope: "error" (default) greps the error payload only; scope: "all" also greps full per-node run data (slower, may return node outputs - treat snippets as sensitive). Optional workflowId, status, limit (default 50, max 250), maxMatches (default 20), snippetChars (default 160). Returns matches plus a skipped array for any execution that failed to fetch.

n8n_list_webhooks - scans workflows for webhook and form-trigger nodes and returns their paths + fully-formed triggerUrl. Pairs with n8n_trigger mode='webhook'. Optional workflowId, activeOnly (default true), limit (default 50).

n8n_validate_workflow - checks for deprecated node types (function → code), legacy Code-node API ($node[], items global, require()), orphan nodes, disabled nodes, missing trigger. Returns issues with severity (error/warning/info) plus a summary count.

n8n_list_schedules - scans n8n-nodes-base.scheduleTrigger and the legacy n8n-nodes-base.cron nodes across workflows and decodes each interval rule into a human-readable string. Answers "what's running at 3am?" without clicking through the n8n UI. Supported rule fields: seconds / minutes / hours / days / weeks / months (with triggerAtHour, triggerAtMinute, triggerAtDay, triggerAtDayOfMonth) and raw cronExpression. One entry per interval - multi-interval rules emit multiple rows. Each row includes workflowId, workflowName, active, nodeName, nodeType, schedule, field, optional cronExpression, and the original raw rule for further inspection. Optional workflowId (single-workflow scan), activeOnly (default true - inactive schedules don't fire), limit (default 100, max 250).

n8n_diff_workflow - compare a workflow's current state against a snapshot. Pass id plus exactly one of snapshotPath or snapshot (inline object). snapshotPath is confined to the configured backupDir (default ~/.n8n-backups): it may be given relative to that directory or as an absolute path inside it, but any path that resolves outside backupDir (including .. traversal) is rejected before the file is read. This keeps the tool from being used as an arbitrary file-read primitive even though it is available without enableEdit. Snapshot accepts both shapes: the flat backup written by n8n_save_workflow / n8n_delete_workflow, and the nested n8n_get_workflow(includeDefinition=true) shape (graph data under definition). Returns summary (counts: added/removed/modified/nameChanged/connectionsChanged/settingsChanged) plus diff with per-node fieldsChanged paths (e.g. parameters.command, parameters.url, disabled). Node matching is two-pass: id first, then name fallback for any unmatched nodes - handles legacy/hand-edited snapshots. Cosmetic changes (position, webhookId) are suppressed by default; pass ignoreCosmetic: false to surface them. Per-node detail is capped at maxModifiedDetails (default 50, max 500); summary.nodesModified counter is uncapped and diff.nodesModifiedTruncated: true flags when detail was clipped. Read-only.

n8n_audit_browser_bridge_usage - scans every workflow for nodes that invoke the browser-bridge CLI. Inspects command (Execute Command + SSH nodes) and jsCode / pythonCode / functionCode (Code + legacy Function nodes). Heuristic: \bbrowser-bridge\.[cm]?js followed by two kebab-slug args; the bare bin form is intentionally not detected to avoid false positives from path mentions like cd /opt/browser-bridge. Returns one finding per (workflowId, nodeName, platform, action) plus a summary of platform×action counts. Optional platform, action, activeOnly (default false), includeArchived (default false), maxWorkflows (default 250, max 1000), concurrency (default 3, max 8). Read-only. Pairs with n8n_scaffold_browser_bridge_node when you need to add another call. Companion repo: browser-bridge.

n8n_scaffold_browser_bridge_node - pure local generator (no n8n API call). Given platform, action, optional input JSON, and mode: "code-node" | "execute-command" (default code-node), emits a ready-to-paste n8n node JSON that mirrors browser-bridge's docs/n8n-usage.md patterns. The Code node uses spawnSync with stdin JSON and surfaces payload.exitCode + stderr so downstream nodes can branch on ok. The Execute Command node uses a quoted <<'JSON' heredoc so the input passes through unmangled. Optional bridgeDir, nodeName, position. Platform/action are validated as kebab slugs - keeps them safe to interpolate into the shell command. Warns when execute-command is used with non-empty input (heredoc bakes the JSON in; no per-item upstream wiring).

n8n_trigger - write tool (hidden unless enableEdit); requires confirm: true. Triggering runs the workflow's nodes (Code / Execute Command / HTTP, etc.) and POSTs to webhooks, all of which can have arbitrary real-world side effects, so it lives behind the same edit gate as the other write tools and refuses without confirm: true. webhookPath is validated client-side: it must resolve to a path under /webhook, /webhook-test, or /form, with no .. traversal or scheme-relative //host form, so a confused agent cannot redirect the call off the base URL. Two modes:

  • mode: "webhook" + webhookPath - POST (or GET/PUT/DELETE) to the configured base URL + path, with an optional JSON payload. This is the reliable path.

  • mode: "workflow" + workflowId - attempts POST /api/v1/workflows/:id/execute. Pre-checks that the workflow is active and has a webhook/manual/form trigger. Most n8n builds don't expose this endpoint on the Public API and will 405; the tool surfaces a hint to switch to webhook mode.

n8n_create_workflow - POST /workflows. Creates from structured workflow JSON: name, nodes, connections, optional settings / staticData. Also accepts the full output of n8n_get_workflow (with includeDefinition=true) and backup snapshots directly. Strips read-only fields (id, active, createdAt, updatedAt, isArchived, versionId, triggerCount, tags, shared, meta, pinData) before POSTing - n8n enforces additionalProperties: false on the workflow schema and will 400 on any readOnly field. Runs n8n_validate_workflow on the proposed state as a pre-check; errors block, warnings pass through (pass skipValidation: true to bypass). Optional dryRun:true returns the cleaned POST body and validation issues without writing (no confirm needed for a dry run). Optional projectId and folderId are sent as create-target query params. Requires confirm: true to actually write - the tool accepts an arbitrary nodes graph that will live on the server, so an unconfirmed call (without dryRun) returns ok: false and never touches the API. The new workflow is created INACTIVE; call n8n_activate afterwards if you want triggers running. This is also the primary restore path for n8n_delete_workflow snapshots: read the backup file into definition and call this tool. The restored workflow gets a new id.

n8n_activate / n8n_deactivate - idempotent; both require confirm: true. Activating arms the workflow's triggers, so its nodes can start running automatically (effectively arming arbitrary-code execution); deactivating halts that automation. Deactivating does not cancel running executions.

n8n_save_workflow - before writing: fetches the current version, snapshots it to backupDir as <id>-<timestamp>.json (mode 0600), runs validateWorkflow on the proposed state, and aborts on error-severity issues (pass skipValidation: true to bypass). Requires confirm: true to actually PUT; calling with confirm: false returns ok: false and never touches the API (omitting confirm is rejected at the MCP schema layer). Response includes the backup path and a restoreHint.

n8n_archive_workflow - POST /workflows/{id}/archive. Soft-deletes a workflow: triggers stop firing, the workflow disappears from the default UI list, but the definition and execution history are preserved. Idempotent (archiving an already-archived workflow returns the current state). Requires confirm: true (it deactivates and soft-deletes the workflow); reversible via n8n_unarchive_workflow. Archiving deactivates as a side effect; the response surfaces active: false explicitly. Returns ok: false with reason: "not_found" on 404.

n8n_unarchive_workflow - POST /workflows/{id}/unarchive. Restores an archived workflow. Does NOT reactivate - triggers stay off until you call n8n_activate explicitly. Requires confirm: true; omitting it or passing false returns ok: false and never touches the API. Returns ok: false with reason: "not_found" on 404.

n8n_delete_workflow - DELETE /workflows/{id}. Permanent, irreversible. Before firing the DELETE: fetches the current workflow and snapshots it to backupDir as <id>-DELETED-<timestamp>.json (mode 0600). If the snapshot can't be written, the DELETE is aborted - there is no un-safety-netted path. Requires confirm: true; omitting it or passing false returns ok: false and never touches the API. Returns ok: false with reason: "not_found" on 404 (either before or after the snapshot). Restore is one-call via n8n_create_workflow with the snapshot contents; the restored workflow gets a new id and is created inactive. Deleting does NOT cancel running executions - use n8n_list_executions(workflowId, status='running') + n8n_cancel_execution first if needed. Prefer n8n_archive_workflow for cleanup if you want to preserve the original id.

n8n_cancel_execution - POST /executions/{id}/stop. Closes the triage loop after n8n_search_executions locates a stuck run. Requires confirm: true; omitting it or passing false returns ok: false and never touches the API. Cancellation can leave a multi-step automation partially applied (e.g. a record written but its follow-up notification or cleanup never run). Returns a success summary with the execution's final status, or ok: false with reason: "not_found_or_finished" if the id no longer matches a running execution (404).

n8n_retry_execution - POST /executions/{id}/retry. Creates a NEW execution - the response surfaces both originalExecutionId and newExecutionId so agents can follow up with n8n_get_execution on the retry. Requires confirm: true; omitting it or passing false returns ok: false and never touches the API. Each retry may re-run side effects (HTTP calls, DB writes, etc). Verify the workflow is safe to re-run before confirming. Optional loadWorkflow: true retries against the currently saved workflow instead of the version captured at original execution time. Returns ok: false with reason: "not_found" on 404 or reason: "not_retryable" on 409 (e.g. still running); all other API errors rethrow.

n8n_delete_execution - DELETE /executions/{id}. Permanently removes an execution record: logs, per-node run data, and error payloads are erased from n8n. Requires confirm: true to actually delete; calling with confirm: false returns ok: false and never touches the API (omitting confirm is rejected at the MCP schema layer). Returns ok: false with reason: "not_found" on 404; all other API errors rethrow. Not idempotent from an agent's perspective: the record is gone after the first successful call, so fetch n8n_get_execution first if you may need it later.

n8n_pin_node_data - pin sample data to a node so downstream nodes use it during testing/development without re-running the upstream node. Pairs naturally with n8n_scaffold_browser_bridge_node: scaffold a browser-bridge call, run it once, capture the output, pin it, then iterate on downstream nodes without re-spawning the browser. Inputs: id, nodeName (case-sensitive, must match an existing node), data (1-50 items; raw objects are auto-wrapped into {json: <object>}, items already shaped as {json: ..., binary?: ...} pass through unchanged), optional merge: true to append to existing pinned data instead of replacing (combined still capped at 50), confirm: true. Issues PUT /workflows/{id} with merged pinData plus the existing nodes/connections/settings/staticData (so the PUT does not blank them). Pinned data persists across executions until cleared - easy to forget; the response includes an unpinHint.

n8n_unpin_node_data - clear pinned data on one node (when nodeName is supplied) or the whole workflow (when omitted). Idempotent: clearing a node that wasn't pinned returns ok: true with noop: true and never touches the API. When clearing actually happens, the PUT includes the rest of the workflow body so other fields are not blanked. Requires confirm: true.

n8n_delete_executions - batch form. Client-side fan-out over DELETE /executions/{id} with bounded concurrency (default 3, max 10). Takes an ids array (deduped before fan-out, capped at 50), requires confirm: true. Response surfaces requested/attempted/deleted/alreadyDeleted/failed/skipped/aborted counters plus a results: Array<{id, ok, reason?, message?}> - order is completion order, not input order, so look up by id. 404 per id is treated as already_deleted (idempotent). A 5xx on any id aborts the batch via an AbortController: no new ids are claimed and any already-in-flight fetches are cancelled client-side. Under concurrency N, up to N-1 deletes may have already reached the server before the 5xx is observed, so the batch is best-effort, not transactional - clear signal the server is sick; don't retry blindly. Per-id error messages are passed through the API-key redactor. Compose with n8n_search_executions to purge a known set of noisy runs in one call.

n8n_retry_executions - batch form of retry. Same fan-out shape as n8n_delete_executions: bounded concurrency (default 3, max 10), capped at 50 ids, AbortController on 5xx, results in completion order. Differs in two ways: 404 per id is { ok: false, reason: "not_found" } (NOT idempotent - a missing execution is a real failure to surface), and each successful retry creates a NEW execution whose id is returned per row as newExecutionId. Counters: requested/attempted/retried/notFound/failed/skipped/aborted. Optional loadWorkflow: true retries every id against the currently saved workflow instead of the captured version. Confirm-gated - each retry runs the workflow again and may re-trigger side effects (HTTP calls, DB writes); verify the workflow is safe to re-run before confirming.

n8n_run_audit - POST /audit. Runs n8n's built-in security audit and returns one risk report per requested category: credentials (unused/abandoned), database (SQL-injection-prone expressions in query nodes), nodes (community/unofficial nodes), filesystem (host fs access from nodes), instance (insecure server settings). Each report has risk, sections (with title/description/recommendation/location). The tool also surfaces a flat reports array with per-report sectionCount/locationCount so an agent can decide what to drill into without reparsing the whole audit. Optional categories (omit for all five) and daysAbandonedWorkflow (n8n default 90). Read-only - n8n only inspects, never mutates. Requires the API user to be an instance admin or owner (n8n's audit endpoint enforces this).

n8n_find_workflows_using_node_type - composed read-only scanner. Walks every workflow (paginated, capped at maxWorkflows, default 250 / max 1000) and emits one finding per node matching the requested type. match: "exact" (default) is full-string equality on node.type; match: "contains" is case-insensitive substring (handy for "all Slack nodes across base + community packages"). Optional activeOnly (default false), includeArchived (default false), includeDisabledNodes (default true - disabled nodes are common drift signals worth surfacing), concurrency (default 3, max 8). Returns per-node findings plus a per-workflow summary sorted by match count descending. Per-workflow fetch errors land in fetchErrors instead of failing the whole scan. Pairs with n8n_audit_browser_bridge_usage (which schedules drive my browser-bridge calls?) and n8n_run_audit (which deprecated nodes need replacing?).

n8n_execution_stats - composed read-only aggregator over n8n_list_executions. Per-workflow counts (total/success/error/canceled/running/waiting/other), failure rate (error / (success + error + canceled)), avg + p95 runtime over completed executions, and lastFailureAt / lastSuccessAt. Optional workflowId (single-workflow stats), sinceHours (default 24, max 168 = 7d), maxExecutions (default 1000, max 5000), pageSize (default 250). Pagination stops on the first execution older than the window OR at maxExecutions; stoppedReason is one of "window", "cap", "exhausted". If truncated: true, increase maxExecutions or narrow sinceHours. The totals object includes the same counts + failureRate rolled across all workflows in the window. Useful for "which workflows are flaky?" and "what's running long?"

n8n_list_tags - GET /tags. Returns { data: [{id, name, createdAt, updatedAt}], nextCursor }. Optional limit (default 100, max 250) and cursor (from a previous call's nextCursor). Read-only.

n8n_get_workflow_tags - GET /workflows/{id}/tags. Returns the array of tag objects currently attached. Pairs with n8n_set_workflow_tags for diffs and reattach flows.

n8n_create_tag - POST /tags. Requires confirm: true (consistent with the other mutating tag tools); reversible via n8n_delete_tag. The name is trimmed before send. Returns ok: false with reason: "conflict" on 409 (tag with this name already exists); use n8n_list_tags to find the existing id.

n8n_delete_tag - DELETE /tags/{id}. Confirm-gated. Cascades: n8n removes the tag from every workflow it was attached to. The workflows themselves are NOT deleted, only the tag association. Returns ok: false with reason: "not_found" on 404. To find affected workflows beforehand, use n8n_list_workflows(tags=<name>) or scan n8n_get_workflow_tags.

n8n_set_workflow_tags - PUT /workflows/{id}/tags. REPLACES the workflow's tag set (not append) - pass the full desired list. Empty tagIds: [] clears all tags. Tag ids are deduped before send. Requires confirm: true (reversible by re-setting). Returns ok: false with reason: "not_found" on 404 (the workflow id OR one of the tag ids does not exist; verify both with n8n_list_workflows and n8n_list_tags).

n8n_list_credentials - GET /credentials. Returns metadata only - n8n's API explicitly excludes the data field (encrypted secrets) from list responses, and the tool layer strips data defensively in case of a future regression. Each row: {id, name, type, createdAt, updatedAt, shared[]}. Optional limit (default 100, max 250) and cursor. Requires the API key to belong to an instance owner or admin - non-admin keys get ok: false, reason: "unauthorized" with a clear hint.

n8n_get_credential_schema - GET /credentials/schema/{credentialTypeName}. Returns the raw JSON Schema describing the required data shape for a credential type (e.g. freshdeskApi → { apiKey, domain } required). Use this before calling n8n_create_credential so you know what fields to populate. 404 returns reason: "not_found"; 401 returns reason: "unauthorized".

n8n_find_workflows_using_credential - composed scanner (no direct n8n endpoint). Walks workflows and inspects every node's credentials field. Pass either credentialId (exact, preferred) or credentialName (case-insensitive substring fallback). Returns one finding per (workflowId, nodeName, credentialType) plus a per-workflow summary count. Same fan-out shape as n8n_audit_browser_bridge_usage (bounded concurrency, fetchErrors for per-workflow failures, truncated flag, maxWorkflows default 250). The answer to "I'm rotating Slack creds, where do I need to update?" - run this before n8n_delete_credential to see the blast radius.

n8n_check_disabled_nodes - composed scanner. Surfaces every node with disabled: true across recent workflows. One finding per (workflowId, nodeName, nodeType) plus per-workflow disabled count, sorted desc. Disabled nodes are common drift signals (frozen mid-debug, forgotten cleanup) and the n8n UI doesn't list them anywhere obvious. Same fan-out + filter shape as the other scanners.

n8n_create_credential - POST /credentials. Double-gated: requires both enableEdit AND enableCredentialsWrite (default false). Confirm-gated. data carries plaintext secrets to n8n; the tool layer never echoes data back, even on error - n8n 400s with body content are wrapped to a status-only error before surfacing, so secrets cannot leak via validation messages. Pre-call: use n8n_get_credential_schema to learn the required data shape. Post-call: response includes id, name, type, timestamps; no data. NOT idempotent - calling twice with the same name creates two credentials.

n8n_delete_credential - DELETE /credentials/{id}. Double-gated + confirm-gated. Cascades: every workflow referencing this credential will fail on its next run - call n8n_find_workflows_using_credential first to enumerate the blast radius. 404 returns reason: "not_found". The deleted-credential payload echoed by n8n has data stripped at the tool layer regardless of upstream behavior.

Configuration

Generate an API key in n8n under Settings → API, then set these env vars in your MCP client config:

Variable

Required

Default

Description

N8N_BASE_URL

yes

-

n8n base URL, e.g. https://n8n.example.com

N8N_API_KEY

yes

-

n8n Public API key (X-N8N-API-KEY)

N8N_ENABLE_EDIT

no

false

Expose write tools

N8N_ENABLE_CREDENTIALS_WRITE

no

false

Second gate (on top of N8N_ENABLE_EDIT) for n8n_create_credential and n8n_delete_credential. See Security model.

N8N_BACKUP_DIR

no

~/.n8n-backups

Where n8n_save_workflow writes pre-save snapshots

N8N_MAX_EXECUTION_LOG_BYTES

no

65536

Cap on inline execution log bytes

N8N_REQUEST_TIMEOUT_MS

no

15000

HTTP timeout for n8n API calls

CLI

n8nctrl is the standard read-only operator control CLI for shells, cron, and CI. It talks to the same n8n Public API as the MCP server and shares the same client core, so what the agent can read, you can read from a terminal. It exposes only the read/report tools - no create, save, archive, delete, cancel, retry, or trigger. The older n8n-ops bin remains a compatibility alias.

# installed globally, the standard bin is `n8nctrl`:
n8nctrl workflows list --active
n8nctrl workflows get <id> --full
n8nctrl workflows validate <id>
n8nctrl executions list --status error --since 24
n8nctrl executions search "ECONNREFUSED" --scope all
n8nctrl executions stats --since 48
n8nctrl webhooks list
n8nctrl schedules list
n8nctrl tags list
n8nctrl credentials find-usage <credentialId>
n8nctrl nodes find n8n-nodes-base.slack --contains
n8nctrl nodes check-disabled
n8nctrl audit run            # exit 1 if the backend is unreachable (cron-friendly)
n8nctrl audit browser-bridge
n8nctrl --json tags list     # JSON for piping

Workflow and execution inspection redacts secret-bearing fields, inline credentials, and the configured API key before human rendering or JSON serialization. workflows get <id> --json and executions get <id> --json provide redacted inspection views. These views cannot export exact secret-bearing workflow definitions or execution data. Execution run data is included by default, with --no-run-data to omit it. Errors from these reads suppress upstream details that could contain raw definitions or execution data. Credential schema properties and metadata commands retain their existing format.

Run n8nctrl help for the full command and flag list. Configure with the same N8N_BASE_URL / N8N_API_KEY env vars as the MCP server (see the table above). Exit codes: 0 success, 1 runtime error (backend unreachable or a call failed), 2 usage error (unknown command/flag or bad value).

Starting the MCP adapter

n8nctrl mcp starts the stdio MCP adapter. The n8n-ops-mcp bin remains supported for existing MCP client configs and package launchers. If a launcher referenced the built file path dist/mcp-server.js directly, point it at dist/mcp-bin.js (or dist/cli.js mcp); launchers that use the n8n-ops-mcp bin name need no change.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "n8n": {
      "command": "npx",
      "args": ["-y", "n8n-ops-mcp"],
      "env": {
        "N8N_BASE_URL": "https://n8n.example.com",
        "N8N_API_KEY": "your-api-key-here"
      }
    }
  }
}

Claude Code

claude mcp add n8n \
  --env N8N_BASE_URL=https://n8n.example.com \
  --env N8N_API_KEY=your-api-key-here \
  -- npx -y n8n-ops-mcp

Add --scope user to make it available from any directory instead of only the current project.

Codex CLI

codex mcp add n8n \
  --env N8N_BASE_URL=https://n8n.example.com \
  --env N8N_API_KEY=your-api-key-here \
  -- npx -y n8n-ops-mcp

Writes the entry to ~/.codex/config.toml under [mcp_servers.n8n]. Verify with codex mcp list.

Cursor / Windsurf / other MCP hosts

Any MCP-compatible client that accepts a stdio command + env will work. Point it at npx -y n8n-ops-mcp (or the globally installed n8n-ops-mcp binary) with N8N_BASE_URL and N8N_API_KEY in the environment.

Hermes Agent reads MCP config from ~/.hermes/config.yaml:

mcp_servers:
  n8n:
    command: "npx"
    args: ["-y", "n8n-ops-mcp"]
    env:
      N8N_BASE_URL: "https://n8n.example.com"
      N8N_API_KEY: "your-api-key-here"

Then reload from inside a session:

/reload-mcp

OpenClaw (first-class plugin)

n8nctrl ships as a first-class OpenClaw plugin through the n8n-ops-mcp package - not an MCP bridge - so it shares the gateway's process, auth profiles, and hooks.

openclaw plugins install clawhub:n8n-ops-mcp

Add the config block to ~/.openclaw/openclaw.json:

{
  "plugins": {
    "entries": {
      "n8n": {
        "enabled": true,
        "config": {
          "baseUrl": "https://n8n.example.com",
          "enableEdit": false
        }
      }
    }
  }
}

Put the API key in your OpenClaw workspace env:

# ~/.openclaw/workspace/.env
N8N_API_KEY=your-api-key-here

Restart the gateway:

systemctl --user restart openclaw-gateway

Config keys: baseUrl, apiKey, apiKeyEnv, enableEdit, enableCredentialsWrite, maxExecutionLogBytes, requestTimeoutMs, backupDir. See openclaw.plugin.json for the full schema and the Security model for the two-gate write design.

If you want to point OpenClaw at a local clone instead of the registry:

{
  "plugins": {
    "allow": ["n8n"],
    "load": {
      "paths": ["/absolute/path/to/n8n-ops-mcp"]
    },
    "entries": {
      "n8n": {
        "enabled": true,
        "config": {
          "baseUrl": "https://n8n.example.com",
          "enableEdit": false
        }
      }
    }
  }
}

Security model

Two flags gate write access, with deliberately different blast radii:

  • enableEdit (default false) - exposes the workflow + execution lifecycle write tools (create/save/archive/delete workflows, activate/deactivate, trigger, cancel/retry/delete executions, pin/unpin node data, tag CRUD). n8n_trigger is gated here too: running a workflow executes arbitrary Code / Execute Command / HTTP nodes and POSTs to webhooks, so it is treated as a write. Mutating tools are confirm-gated (including trigger, activate, deactivate, archive, create_workflow, create_tag, and set_workflow_tags), and the destructive workflow ones snapshot to backupDir first. The read-only n8n_diff_workflow confines snapshotPath reads to backupDir so it cannot be turned into an arbitrary file-read primitive.

  • enableCredentialsWrite (default false) - second gate, on top of enableEdit, required to expose n8n_create_credential and n8n_delete_credential. An agent that has been overprovisioned with enableEdit cannot inject or destroy credentials without this separate, deliberate config change.

Both flags must be true for credential writes to register. The credential read tools (list-credentials, get-credential-schema, find-workflows-using-credential) and the disabled-node scanner are always available regardless.

Why credentials get a second gate:

  1. create-credential is the only tool in this package where agent input contains plaintext secrets. A prompt-injected or confused agent with enableEdit shouldn't be able to inject credentials.

  2. delete-credential cascades - every workflow referencing the credential fails on its next run. The blast radius is wider than any single workflow operation.

Defense-in-depth on data:

  • n8n's OpenAPI marks data as writeOnly - the API contract excludes it from every response. We trust but verify: the tool layer strips data from every credential response before surfacing, including success paths and the deleted-credential echo, so a future n8n regression can't leak secrets through us.

  • On create-credential errors, the n8n response body (which can echo back fragments of submitted data on validation 400s) is replaced at the client layer with a status-only error message. The tool surfaces status + path only. Tests assert no portion of a forced-400 request body reaches the tool response.

Example prompts

What n8n workflows broke today?

Calls n8n_list_executions with status=error, then n8n_get_execution for the failing run.

Which workflow errored with "ECONNREFUSED"?

Calls n8n_search_executions with query: "ECONNREFUSED".

Trigger the "nightly intel" workflow (requires N8N_ENABLE_EDIT=true)

Calls n8n_list_webhooks to find the path, then n8n_trigger with mode=webhook and confirm: true.

What's running at 3am?

Calls n8n_list_schedules, then filters the result for any schedule whose description contains "03:00" (or whose cronExpression matches an early-morning hour).

What changed in my "intel pipeline" workflow since yesterday's backup?

Calls n8n_diff_workflow with id and snapshotPath pointing to the backup file. Returns added/removed/modified nodes with parameter-level field paths.

Audit my workflows for deprecated Code-node API usage

Calls n8n_list_workflows then n8n_validate_workflow per id, filters for code-node-old-node-ref and code-node-items-global warnings.

Which workflows are flaky this week?

Calls n8n_execution_stats with sinceHours: 168, then sorts by failureRate.

I'm rotating Slack credentials - where do I need to update?

Calls n8n_find_workflows_using_credential with the credential id to enumerate the blast radius before you touch anything.

Deactivate the "experimental-bot" workflow (requires N8N_ENABLE_EDIT=true)

Calls n8n_list_workflows with a name filter, then n8n_deactivate with confirm: true on the matching id.

Kill the execution stuck on ECONNREFUSED (requires N8N_ENABLE_EDIT=true)

Calls n8n_search_executions with query: "ECONNREFUSED", then n8n_cancel_execution with confirm: true on the match.

Purge the noisy test-run execution logs from last week (requires N8N_ENABLE_EDIT=true)

Calls n8n_search_executions to find the ids, then n8n_delete_executions with confirm: true to purge up to 50 in one call. Deletion is irreversible.

Archive the old "staging-bot" workflow - I might need it back someday (requires N8N_ENABLE_EDIT=true)

Calls n8n_list_workflows with a name filter, then n8n_archive_workflow with confirm: true on the match. Reversible via n8n_unarchive_workflow with confirm: true (you'll still need n8n_activate with confirm: true to turn triggers back on).

Delete the abandoned "poc-scraper" workflow - it's been dead for months (requires N8N_ENABLE_EDIT=true)

Calls n8n_list_workflows to find the id, then n8n_delete_workflow with confirm: true. A snapshot lands in backupDir first; restore is one-call via n8n_create_workflow with the snapshot. Prefer n8n_archive_workflow if you want to preserve the original id.

Why not the bigger n8n MCP projects?

n8nctrl is deliberately narrow. It is for operating the n8n you already run, not for authoring new flows from a node catalog.

  • For a catalog/docs tool that indexes n8n's node library so an agent can scaffold new workflows from node metadata, see n8n-mcp. That is the right tool when you want the agent to know about every node type and its parameters.

  • n8nctrl is the right tool when you want the agent or CLI to answer "what's broken, what changed, what's risky, what's scheduled" against your live instance and to act on it: triage failed executions, diff against backups, scan for security and credential blast radius, and edit workflows behind explicit, snapshotted write gates.

The two are complementary. One helps build; this one helps run.

What n8nctrl is not

  • Not a node-catalog or workflow-authoring assistant. It does not index n8n's node library or suggest node parameters. Use n8n-mcp for that.

  • Not a replacement for the n8n UI or its REST API. It is a thin, opinionated operator layer over the n8n Public API, focused on ops questions and safe writes.

  • Not a hosted service. No daemon, no telemetry. It runs as a local stdio process (or an in-process OpenClaw plugin) and talks only to the n8n instance you configure.

  • Not a credential exfiltration path. Credential reads return metadata only; credential writes are double-gated and data is stripped from every response branch. See the Security model.

Development

npm install
npm run dev       # tsx on mcp-server.ts (MCP stdio)
npm run typecheck
npm test          # vitest run
npm run build     # tsup bundle to dist/mcp-server.js
npm start         # node dist/mcp-server.js (post-build)

Or install from source:

git clone https://github.com/lidless-labs/n8nctrl.git
cd n8nctrl
npm install
npm run build

Contributing

Patches welcome. See CONTRIBUTING.md for what lands easily, SECURITY.md for how to report a vulnerability, and the Code of Conduct.

Changelog

See CHANGELOG.md for the full version history.

License

MIT. See LICENSE.


HTTP mode

Run a shared Streamable HTTP server with node dist/mcp-bin.js --http. The existing node dist/mcp-bin.js command continues to use stdio.

Set N8N_BASE_URL, N8N_API_KEY (or the variable selected by N8N_API_KEY_ENV), and an independent random N8N_MCP_HTTP_TOKEN. Startup rejects blank HTTP tokens and tokens equal to the API key or URL credentials. Bearer tokens use SHA-256 digests and constant-time comparison.

Variable

Default

Purpose

N8N_MCP_HTTP_TOKEN

Required

Independent bearer token, at least 32 characters

N8N_MCP_HTTP_HOST

IPv4 loopback

Specific listener bind host; wildcard addresses are refused

N8N_MCP_HTTP_PORT

3117

TCP port, or 0 for an available port

The resolved address is printed to stderr. /mcp requires bearer authentication on every POST, GET, and DELETE. GET /healthz returns only {"status":"ok"} without authentication. Host checks reject DNS rebinding. Supplied Origins must be HTTP or HTTPS loopback origins. Bodies are limited to 1 MiB, sessions to 128, and idle sessions expire after 30 minutes on a minute timer. Active POST calls are excluded from idle expiry. SIGTERM and SIGINT close sessions and the listener.

Configuration, including N8N_ENABLE_EDIT, N8N_ENABLE_CREDENTIALS_WRITE, and N8N_BACKUP_DIR, is read once at startup and used by every session. Credential writes require both write gates. Every mutation still requires confirm: true on that call. Each session has its own MCP server, lazy API client, and tool instances. There are no pending approval tokens or shared tool-visibility sets. HTTP mutation calls are serialized across sessions so pin/unpin read-modify-write operations cannot overwrite another HTTP call's update. HTTP save/delete backups use exclusive writes and UUID suffixes to prevent filename collisions. The backup root and file-read confinement stay unchanged. Other processes and direct n8n clients can still race with this process's writes.

All clients holding the bearer token share the same n8n account and privileges. Keep the default loopback binding. Remote access needs a trusted TLS proxy or tunnel and network access controls. This listener provides no TLS or per-client accounts. Preserve an allowed Host header through a proxy. Clients must initialize again after expiry or restart.

Claude Code

Add this entry under mcpServers in your client configuration:

{
  "mcpServers": {
    "n8n": {
      "type": "http",
      "url": "http://localhost:3117/mcp",
      "headersHelper": "/path/to/helper N8N_MCP_HTTP_TOKEN"
    }
  }
}

The executable helper reads the named environment variable and prints {"Authorization":"Bearer ..."}. For example, save this as the helper and make it executable. Keep its environment private:

#!/usr/bin/env python3
import json
import os
import sys
print(json.dumps({"Authorization": "Bearer " + os.environ[sys.argv[1]]}))

Codex

Export N8N_MCP_HTTP_TOKEN in the environment launching Codex and configure:

[mcp_servers.n8n]
url = "http://localhost:3117/mcp"
bearer_token_env_var = "N8N_MCP_HTTP_TOKEN"

The client needs only the HTTP token. Upstream credentials stay on the server.

systemd user service

See the example unit. Adjust its generic installation paths and build the project there. Create the EnvironmentFile at %h/.config/n8n-ops-mcp/http.env with mode 0600, containing:

N8N_BASE_URL=https://n8n.example.com
N8N_API_KEY=REPLACE_WITH_UPSTREAM_KEY
N8N_MCP_HTTP_TOKEN=REPLACE_WITH_INDEPENDENT_RANDOM_TOKEN
N8N_MCP_HTTP_PORT=3117
N8N_ENABLE_EDIT=true
N8N_ENABLE_CREDENTIALS_WRITE=false
N8N_BACKUP_DIR=/var/lib/n8n-ops-mcp/backups

The backup directory must be writable by the service user. The unit sets NoNewPrivileges=true, PrivateTmp=true, and UMask=0077. Installing or starting it is a separate operator action.

Available Tools

20 tools
n8n_audit_browser_bridge_usageA

Scan every workflow for nodes that invoke the browser-bridge CLI (Execute Command, Code/Function, or SSH nodes). Returns one finding per (workflowId, nodeName, platform, action) so you can answer 'where am I calling Linktree sync from?' without grepping the n8n DB. Read-only. Heuristic: matches browser-bridge[.js|.cjs] <platform> <action> in command/jsCode strings, including spawn-array form.

Inputs: action, platform, activeOnly, concurrency, +2 more.

n8n_check_disabled_nodesA

Scan workflows and surface every node with disabled: true. One finding per (workflowId, nodeName, nodeType) plus a per-workflow count. Read-only. Disabled nodes are common drift signals — frozen mid-debug, forgotten cleanup — and the n8n UI doesn't surface them in any list view. Bounded-concurrency fan-out; per-workflow fetch errors land in fetchErrors instead of failing the scan.

Inputs: activeOnly, concurrency, maxWorkflows, includeArchived.

n8n_diff_workflowA

Compare a workflow's current state against a snapshot (file path or inline object). Returns a structured semantic diff: nodes added/removed/modified (with per-node field paths), plus name/connections/settings change flags. Snapshot accepts both n8n_save_workflow backup shape (flat) and n8n_get_workflow(includeDefinition=true) shape (nested). Read-only.

Inputs: id (required), snapshot, snapshotPath, ignoreCosmetic, +1 more.

n8n_execution_statsA

Aggregate execution stats over a recent window. Computes per-workflow counts (total/success/error/canceled/running/waiting), failure rate, avg + p95 runtime, last failure + last success timestamps. Composed read-only — paginates /executions and stops on the window boundary or maxExecutions. Useful for 'which workflows are flaky?' and 'what's running long?'. Pagination is best-effort: if truncated: true, increase maxExecutions or narrow sinceHours.

Inputs: pageSize, sinceHours, workflowId, maxExecutions.

n8n_find_workflows_using_credentialA

Scan workflows and surface every node that references a given credential. Pass either credentialId (exact, preferred) or credentialName (case-insensitive substring fallback). Returns one finding per (workflowId, nodeName, credentialType) plus a per-workflow summary count. Read-only. Bounded-concurrency fan-out; per-workflow fetch errors land in fetchErrors instead of failing the whole scan. Pairs with n8n_run_audit and is the answer to 'I'm rotating creds, where do I need to update?' before calling n8n_delete_credential.

Inputs: activeOnly, concurrency, credentialId, maxWorkflows, +2 more.

n8n_find_workflows_using_node_typeA

Scan workflows and surface every node matching a given type (e.g. 'n8n-nodes-base.slack'). Returns one finding per matching node + a per-workflow summary so agents can answer 'where am I calling Slack?' or 'which workflows still use the legacy HTTP Request node?' without grepping the n8n DB. Read-only. Bounded-concurrency fan-out; per-workflow fetch errors land in fetchErrors instead of failing the whole scan.

Inputs: nodeType (required), match, activeOnly, concurrency, +3 more.

n8n_get_credential_schemaA

Fetch the JSON schema for a credential type via GET /credentials/schema/{credentialTypeName}. Returns the raw JSON Schema document describing the required data shape (e.g. freshdeskApi requires { apiKey, domain }). Use this BEFORE calling n8n_create_credential so you know what fields to populate. 404 on unknown type returns { ok: false, reason: 'not_found' }. 401 surfaces the admin/owner role requirement.

Inputs: credentialTypeName (required).

n8n_get_executionA

Fetch a single n8n execution by id. Returns status, mode, timing, and per-node run data. Large run logs are truncated with a tail hint. Error executions include the raw error verbatim.

Inputs: id (required), includeRunData.

n8n_get_workflowA

Fetch a single n8n workflow by id. Returns metadata and optionally the full node graph. Resolves against the live workflow, not the workflow_entity row directly.

Inputs: id (required), includeDefinition.

n8n_get_workflow_tagsA

Read the tags currently attached to a workflow via GET /workflows/{id}/tags. Returns the array of {id, name} tag objects (also includes createdAt/updatedAt). Read-only. Pairs with n8n_set_workflow_tags for diffs and reattach flows.

Inputs: id (required).

n8n_list_credentialsA

List credentials via GET /credentials. Returns metadata only — n8n's API explicitly excludes the data field (the encrypted secrets). Each row: {id, name, type (e.g. 'githubApi'), createdAt, updatedAt, shared[]}. Read-only. Requires the API user to be an instance owner or admin — non-admin keys get 401 with a clear hint. The tool defensively strips any data field that might appear in a future regression.

Inputs: limit, cursor.

n8n_list_executionsA

List recent n8n executions with optional filters. Returns id, workflowId, workflowName, status, mode, startedAt, stoppedAt. Use n8n_get_execution for the full run log.

Inputs: limit, status, workflowId.

n8n_list_schedulesA

Surface every schedule trigger across workflows so you can answer 'what's running at 3am?' without clicking through the n8n UI. Walks scheduleTrigger and legacy cron nodes, decodes their interval rules into human-readable strings (e.g. 'every 2 hours', 'daily at 03:00', 'cron: 0 */6 * * *'), and returns workflow context + the raw rule for further inspection. Read-only.

Inputs: limit, activeOnly, workflowId.

n8n_list_tagsA

List workflow tags via GET /tags. Returns { data: [{id, name, createdAt, updatedAt}], nextCursor }. Read-only; pairs with n8n_set_workflow_tags + n8n_get_workflow_tags for cross-cutting workflow metadata.

Inputs: limit, cursor.

n8n_list_webhooksA

Surface webhook and form-trigger paths from n8n workflows so agents can call n8n_trigger with mode='webhook' without opening the n8n UI. Returns workflowId, workflowName, nodeName, method, path, and a fully-formed triggerUrl.

Inputs: limit, activeOnly, workflowId.

n8n_list_workflowsA

List n8n workflows with optional filters. Returns id, name, active state, tags, updatedAt. Use n8n_get_workflow to pull the full definition.

Inputs: name, tags, limit, active.

n8n_run_auditA

Generate n8n's built-in security audit via POST /audit. Returns one risk report per requested category: credentials (unused/abandoned), database (SQL injection-prone expressions), nodes (community/unofficial nodes), filesystem (host fs access), instance (insecure server settings). Each report has risk, sections (with title/description/recommendation/location). Read-only — n8n only inspects, never mutates. Requires the API user to be an instance admin or owner.

Inputs: categories, includeDetails, daysAbandonedWorkflow.

n8n_scaffold_browser_bridge_nodeA

Generate a ready-to-paste n8n node JSON that calls the browser-bridge CLI for a given (platform, action, input). Mirrors the patterns in browser-bridge's docs/n8n-usage.md so n8n workflows don't need to rediscover the spawn/heredoc shape every time. Pure local generator — no n8n API call. Default mode 'code-node' uses spawnSync with stdin JSON; 'execute-command' uses an Execute Command node with a quoted heredoc.

Inputs: action (required), platform (required), mode, input, +3 more.

n8n_search_executionsA

Text-search recent n8n executions without paging through them one by one. Fetches each candidate with includeData=true, then greps the error payload (scope='error', default) or the full per-node run log (scope='all'). Returns matched executions with workflow context and a snippet around each hit. Snippets are raw run data with the API key redacted — with scope='all' they may still contain credentials or payload data from node outputs, so treat as sensitive.

Inputs: query (required), limit, scope, status, +3 more.

n8n_validate_workflowA

Static checks on a workflow: deprecated node types (function → code), old Code-node API usage ($node[], items global, require()), orphan nodes, disabled nodes, and missing trigger. Returns a list of issues with severity.

Inputs: id (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.1/5.0

    Scored across 20 tools

    Disambiguation5/5

    Every tool has a clearly distinct purpose. For example, execution-related tools (n8n_execution_stats, n8n_get_execution, n8n_list_executions, n8n_search_executions) each focus on different aspects: aggregated stats, single execution detail, listing, and text search. Similarly, workflow scanning tools target different signals (disabled nodes, credential usage, node type usage). No two tools overlap in function.

    Naming Consistency5/5

    All tool names follow the consistent pattern `n8n_<verb>_<resource>` using snake_case and imperative verbs (e.g., list, get, check, diff, find, run, scaffold, validate). The convention is uniform and predictable, making it easy for an agent to infer purpose from the name.

    Tool Count4/5

    With 20 tools, the server is on the higher end but still appropriate for the broad domain of n8n operations (workflows, executions, credentials, tags, schedules, webhooks, auditing, node generation). Each tool provides distinct value, and the count reflects the complexity of the platform without being excessive.

    Completeness3/5

    The tool set covers inspection, auditing, validation, and comparison well, but lacks mutation tools for creating, updating, or deleting workflows and credentials. While the server's focus appears to be ops/inspection, the absence of basic CRUD operations for key resources creates notable gaps that may require agents to fall back to other means.

    Maintenance

    ActivityNo data
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      C
      maintenance
      Enables management of n8n workflows, executions, credentials, tags, and variables via MCP tools. Supports stdio, Claude Code, and Claude Web with optional Authentik OAuth.
      22
      MIT
    • F
      license
      B
      quality
      C
      maintenance
      Enables AI assistants to fully automate n8n instances via 34 MCP tools, covering workflow creation, execution debugging, credential management, environment variables, tags, and user administration through natural language.
      34
      1
      -
    • A
      license
      A
      quality
      C
      maintenance
      Lets AI agents read, create, and update n8n workflows with a guarded two-phase plan-and-apply flow that backs up every change, refuses writes when the live workflow has moved, and verifies results by readback. Also supports rehearsing changes on trigger- and outbound-disabled test clones, so mutations to a live instance stay reviewable and reversible.
      5
      27 npm
      6
      MIT