mrkeyoor.com_
Wed 16 Sept 23:57 UTC
AI Toolsevaluationupdated 26 Aug 2026

n8n-mcp review

n8n-MCP is a Model Context Protocol server that lets AI assistants look up n8n nodes, inspect templates, validate workflows, and manage an n8n instance. It solves the tedious part of giving an assistant accurate node schemas and operations instead of asking it to invent workflow JSON from memory.

+42stars / 7d
Verdict

Our n8n-MCP run installed 2,761 packages and passed 6,387 of 6,660 tests, while 220 failed and npm audit reported 133 known vulnerabilities. That makes it useful for assisted discovery and drafting, but a poor candidate for unsupervised production writes. Use it with read-only access first, then enable narrowly chosen write operations only after full workflow diffs and backups are routine.

We ran it

Lab card: what happened when we ran n8n-mcpScreenshot of n8n-mcp (www.n8n-mcp.com)
Install✓ · 152s2761 packages · 1775 MB
Build✓ · 12s
Tests✗ · 706s6387 passed · 220 failed · 53 skipped of 6660 (vitest)
Known vulns1335 critical · 42 high · 82 moderate · 4 low (npm audit)
Repo884 files~235,128 lines of source · 116 MB · 8 CI workflows · Dockerfile · tests dir

Answers from our run

Does n8n-mcp build from source?

Dependencies installed in 152 seconds (2761 packages), and the build succeeded in 12 seconds. We cloned commit fe3a4cb into a clean Debian container with 3 CPUs and no project-specific setup.

Do n8n-mcp's tests pass?

Not all of them: 6387 of 6660 passed and 220 failed when we ran the project's own test command (vitest). Some failures need services or credentials a bare container does not have.

Does n8n-mcp have known vulnerabilities in its dependencies?

npm audit flagged 133 known advisories in the dependency tree, including 5 critical at the time of our run.

Who should not use n8n-mcp?

Anyone planning to let an agent edit production workflows unattended: the README explicitly says to copy, back up, test, and validate workflows first.

What are the alternatives to n8n-mcp?

n8n, Activepieces, Node-RED. Our n8n-MCP run installed 2,761 packages and passed 6,387 of 6,660 tests, while 220 failed and npm audit reported 133 known vulnerabilities.

Setup2/51,775 MB install and 220 failing tests raise the trial cost
Docs5/5Tool, client, deployment, safety, and read-only paths are documented
Community4/522,785 stars, an August push, and active issue discussion
Maturity3/5Broad workflow tooling, with failing tests and open write-path reports

Who it’s for

n8n users who want Claude Code, Cursor, Codex, or another MCP client to search node documentation and templates.
Automation teams willing to review generated workflow changes before they reach a live n8n instance.
Self-hosters who can isolate an MCP service and issue a restricted n8n API key.
Developers who need validation and diff-based edits instead of pasting whole workflows into a chat.

Who it’s NOT for

Anyone planning to let an agent edit production workflows unattended: the README explicitly says to copy, back up, test, and validate workflows first.
Security teams unwilling to expose n8n metadata or credentials through an agent tool: management needs an n8n API key, and some tools can read credentials or data tables unless disabled.
Teams that require a clean dependency audit before evaluation: our install found 133 known vulnerabilities, including 5 critical and 42 high.
Buyers who need every checked-in test passing on the documented commit: our suite ended with 220 failures, and the log shows failures in workflow API integration paths.
MCP clients that require confirmed conformance with the newest advertised revisions: open issues 1019 and 1020 report negotiation and method violations.

Setup reality

Our sandbox installed 2,761 npm packages in 152 seconds and used 1,775 MB. The build passed in 12 seconds. Tests failed after 706 seconds: Vitest reported 6,387 passed, 220 failed, and 53 skipped out of 6,660. Npm audit found 133 known vulnerabilities, including 5 critical and 42 high.

Documentation and validation work without an n8n connection. The 16 management tools need N8N_API_URL and N8N_API_KEY; Cloudflare Access adds a client ID and secret. Hosted use needs an n8n-MCP account and its API key, while self-hosting can use npx, Docker, Compose, or Railway.

Write tools can create, replace, update, and permanently delete workflows. The README provides disabled-tool lists and recommends a read-only n8n key. Treat those controls as setup requirements, since the project's own warning says AI should never edit production workflows directly.

It gives an agent n8n schemas instead of guesses

n8n-MCP puts n8n node documentation, templates, validation, and instance management behind Model Context Protocol tools. An assistant can search for a node, inspect properties and operations, compare versions, find templates, and validate a workflow before deployment. The README describes 7 core discovery tools and 16 management tools. That split matters: the first group can help draft and check workflow JSON, while the second group can change an attached n8n instance.

The catalog is unusually broad. The README lists 2,541 nodes, including 832 core nodes and 1,709 community nodes, plus 2,352 workflow templates. Those are project claims, so they describe coverage rather than proof that every configuration will run. Still, schema lookup is a better input to an agent than remembered examples from a model. Teams already using n8n will get more value than teams shopping for an automation platform.

Read-only mode is the sensible first deployment

Management access uses N8N_API_URL and N8N_API_KEY. Once enabled, tools can create workflows, replace them, apply partial updates, run tests, delete executions, manage folders, and work with credentials or data tables. That is enough power to turn a drafting assistant into an operator. Cloudflare Access installations may also provide a service-token pair, which the server confines to the configured n8n origin.

The README supplies 9 whole-tool exclusions for a read-only deployment, including workflow writes, test execution, credentials, and data tables. A second setting blocks destructive operations inside mixed read and write tools. Pairing those lists with a read-only n8n API key gives two layers of restriction. This is the correct starting posture. The project's own safety warning tells users to copy workflows, test in development, export backups, and validate changes before production deployment.

What happened when we ran it

Our sandbox installed 2,761 npm packages in 152 seconds, leaving 1,775 MB on disk. The TypeScript build completed successfully in 12 seconds. This was a 116 MB checkout with 884 files and roughly 235,128 lines of source, so a quick npx launch hides a substantial dependency and code footprint. The repository includes 8 CI workflow files, a Dockerfile, a Compose file, and a tests directory.

The test command failed after 706 seconds. Vitest reported 6,387 passed, 220 failed, and 53 skipped out of 6,660 across 235 test files. The tail shows Axios calls through N8nApiClient.sendWorkflowWrite and createWorkflow, ending in validate-workflow.test.ts; it does not establish one cause for all 220 failures. We can say the checked-out commit did not pass its full suite in our fresh, secret-free container.

Npm audit found 133 known vulnerabilities: 5 critical, 42 high, 82 moderate, and 4 low. That result concerns the dependency tree installed for commit fe3a4cb, not a claim that each advisory is remotely exploitable through this server. It is still too large to wave away. Before placing the service near credentials or workflow data, operators should inspect the affected packages, reachable paths, and available upgrades.

A nested edit report makes full diffs mandatory

Open issue 1023 describes a 14-node workflow where setNodeParameter on one nested path created a duplicate structure and removed discriminator fields from other nodes. The reporter says the first response carried no validation warnings; fetching and diffing the full workflow revealed the damage. A replacement-style parameter update worked around it. This is one report, yet it concerns the exact partial-write behavior that makes the server attractive.

That report turns the README's safety advice into an operating rule. Keep a versioned export, apply changes to a copy, fetch the complete result, and compare every node before promotion. Validation alone may miss a bad write if the write response does. For governance-sensitive installations, leave update and delete tools disabled until the team has an approval step outside the agent conversation.

Protocol and validation reports deserve a trial on your client

Issue 1020 reports that a client offering MCP revision 2025-11-25 received revision 2025-03-26, outside the advertised supported window. A related issue covers 7 findings against revision 2026-07-28. These automated reports may expose compatibility gaps specific to a client and protocol revision, so test the exact client-server pair you intend to deploy. A successful tool listing does not exercise pagination, unknown-tool errors, resources, or every negotiated capability.

Another open report, issue 1018, describes intermittent connection closures during validation of a 9-node workflow. The same report says small read-only calls continued working and a 2-node synthetic workflow did not reproduce the behavior. We did not reproduce that issue in our lab, and our own failures provide no basis for linking the two. Together they justify a staged evaluation built around representative workflows, rather than a toy trigger connected to one action.

August maintenance is active, while the risk stays operational

GitHub recorded the last push on August 24, 2026, and listed 63 combined issues and pull requests. The latest release, v2.73.0, shipped on August 19 and updated n8n compatibility to the 2.35 line. Recent issue updates on August 23 and 24 cover validation, nested edits, and MCP conformance. That is active maintenance and active scrutiny, not an empty tracker.

For documentation search and workflow drafting, n8n-MCP solves a real context problem. For production control, its 16 management tools enlarge the blast radius of an assistant mistake. The 1,775 MB install, 220 failed tests, security-audit count, and open write-path report outweigh the convenience of immediate write access. Start with search and validation, keep the API key restricted, and require a human-reviewed full diff before any workflow changes leave a development instance.

Alternatives

ProjectWhat it isPick it when
n8n gh↗The automation platform itself, with its own editor and AI-oriented workflow features.pick this instead when a human will build in the visual editor and an external MCP control layer adds unnecessary access.
Activepieces gh↗A self-hosted automation platform with a visual builder and extensible pieces.pick this instead when you are still choosing the automation platform rather than adding an agent interface to n8n.
Node-RED gh↗A flow-based automation tool with a mature visual editor and broad device integrations.pick this instead when local event flows and direct human editing matter more than AI-generated n8n workflows.

Sources

  1. n8n-MCP README
  2. n8n-MCP repository activity
  3. Release v2.73.0
  4. Issue 1023: nested workflow update corruption report
  5. Issue 1020: MCP revision negotiation report
  6. Issue 1018: intermittent validation closure report

More ai tools reviews

eve · MemOS · LongCat-Video · Concat · DLSS5-Feeder · Concat · the whole board →