InterviewStack.io MCP server

InterviewStack.io Job Search - for your AI assistant

Search live, curated jobs from the InterviewStack.io job board from any MCP-compatible AI assistant - Claude Code, Cursor, VS Code + Copilot, OpenAI Codex, Windsurf, and more. Let your AI find roles that genuinely fit you, polish your resume against real postings, and draft tailored applications.

Under the hood this is one open MCP server - the same endpoint and tools for every client. This repo packages it for easy setup plus the workflow guidance that makes it sing: a Claude Code plugin (one-click, bundles the skills) and a portable AGENTS.md that Codex, Cursor, Copilot, Windsurf and ~20 other tools read natively.

Jobs are sourced from InterviewStack.io (the .io job board - not a .com).

---

Quick start - one copy-paste, then just talk

1. Get a key (free; see Free vs Pro for what each plan returns) - sign in at app.interviewstack.io/sidenav/job-search-mcp and click Create key. That page fills your key into the command below automatically and has a Test connection button to confirm it works. One active key at a time - delete it from the same page if you need a fresh one.

2. Connect Claude Code - paste this ONE line in your Terminal (your key is embedded, no shell setup, nothing to forget):

   claude mcp add --transport http interviewstack-jobs https://mcp-job-search.interviewstack-io.workers.dev/mcp --header "Authorization: Bearer isk_your_key_here"
   

   Then fully quit and reopen Claude Code, and run /mcp to confirm interviewstack-jobs is connected. Using a different tool? See Step 2.

3. (Optional) add the guided skills - paste in Claude Code for auto-updating tailor/resume/digest workflows:

   /plugin marketplace add interviewstackio/interviewstack-jobs
   /plugin install interviewstack-jobs@interviewstack-jobs
   

4. That's the whole setup. Now just ask:

   "Find new jobs that match my resume and save the best ones for me every morning."

   Your assistant takes it from there: asks for your resume, sets up the morning schedule itself (jittered, on your machine), and from then on the best new matches land in your application tracker with a note on why each one fits. You review and apply. Jobs you remove are never re-saved. More ideas: starter prompts.

Prefer to script the schedule yourself instead of asking in chat? See Step 3.

---

Step 1 - Get your API key (all tools)

Sign in at app.interviewstack.io/sidenav/job-search-mcp and click Create key. Copy the key (isk_…) - it's shown once. The page also has a Test connection button: paste your key there to confirm it works before you wire it into a tool. One active key at a time: delete it from the same page if you need a fresh one (anything still using the old key stops working).

Paste the key directly into each tool's config (shown in Step 2). The configs below embed the literal key - no shell environment variables, which GUI-launched editors often don't inherit and are the #1 cause of a silent "Unauthorized" later. Never commit your key. It's read-only and rate-limited, but treat it like a password; delete-and-recreate anytime.

Step 2 - Connect your tool

The MCP endpoint is the same everywhere: https://mcp-job-search.interviewstack-io.workers.dev/mcp with header Authorization: Bearer <your key>.

Most reliable - one line in your Terminal (key embedded, no env vars):

claude mcp add --transport http interviewstack-jobs https://mcp-job-search.interviewstack-io.workers.dev/mcp --header "Authorization: Bearer isk_your_key_here"

Quit and reopen Claude Code, then verify with /mcp.

Optional - also get the bundled, auto-updating skills (tailor-application, resume-polish, daily-job-digest). Paste inside Claude Code:

/plugin marketplace add interviewstackio/interviewstack-jobs
/plugin install interviewstack-jobs@interviewstack-jobs

The plugin reads your key from an INTERVIEWSTACK_MCP_KEY environment variable, so if you use the plugin route, add export INTERVIEWSTACK_MCP_KEY="isk_…" to your ~/.zshrc (or ~/.bashrc) and relaunch Claude Code from a fresh terminal. The one-line command above needs none of that. Enable auto-update under /plugin → Marketplaces → interviewstack-jobs → Enable auto-update.

Add to ~/.cursor/mcp.json (global - works in every project). Paste your real key in place of isk_your_key_here:

{
  "mcpServers": {
    "interviewstack-jobs": {
      "url": "https://mcp-job-search.interviewstack-io.workers.dev/mcp",
      "headers": { "Authorization": "Bearer isk_your_key_here" }
    }
  }
}

Save the file, then fully quit and reopen Cursor; check Settings → MCP for a green dot. For the workflow guidance, copy AGENTS.md into your project root (Cursor reads it), or add it under .cursor/rules/.

Add to .vscode/mcp.json:

{
  "servers": {
    "interviewstack-jobs": {
      "type": "http",
      "url": "https://mcp-job-search.interviewstack-io.workers.dev/mcp",
      "headers": { "Authorization": "Bearer ${input:interviewstack_key}" }
    }
  },
  "inputs": [
    {
      "id": "interviewstack_key",
      "type": "promptString",
      "description": "InterviewStack.io MCP key",
      "password": true
    }
  ]
}

VS Code prompts once for the key and stores it securely. For guidance, drop AGENTS.md in the repo root - Copilot reads it (or .github/copilot-instructions.md).

Add to ~/.codex/config.toml. The first line (at the very top of the file, not inside any [section]) lets Codex use a literal key instead of an environment variable. Paste your real key in place of isk_your_key_here:

experimental_use_rmcp_client = true

[mcp_servers.interviewstack-jobs]
url = "https://mcp-job-search.interviewstack-io.workers.dev/mcp"
bearer_token = "isk_your_key_here"

Save, restart Codex, then run codex mcp list to confirm. Codex reads AGENTS.md natively - copy it into your project root.

Add to ~/.codeium/windsurf/mcp_config.json. Paste your real key in place of isk_your_key_here:

{
  "mcpServers": {
    "interviewstack-jobs": {
      "serverUrl": "https://mcp-job-search.interviewstack-io.workers.dev/mcp",
      "headers": { "Authorization": "Bearer isk_your_key_here" }
    }
  }
}

Save, then in Windsurf open the Cascade panel → MCP and click Refresh (or restart Windsurf). Windsurf reads AGENTS.md - copy it into your project root.

Point your client at a remote Streamable-HTTP MCP server:

{
  "mcpServers": {
    "interviewstack-jobs": {
      "type": "http",
      "url": "https://mcp-job-search.interviewstack-io.workers.dev/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY_HERE" }
    }
  }
}

Some clients (e.g. older Claude Desktop builds) only support stdio MCP - bridge with mcp-remote: npx mcp-remote https://mcp-job-search.interviewstack-io.workers.dev/mcp --header "Authorization: Bearer YOUR_KEY". Then add AGENTS.md for the workflow guidance.

Once connected, grab a prompt from Starter prompts below.

Not working? Unblock it

Almost every problem is the key. The fastest fix is the Test connection button on the key page - paste your key and it tells you immediately if the key itself is good.

• "Unauthorized" / 401 - the key is missing, mistyped, or you pasted the placeholder (isk_your_key_here) instead of your real key. Copy the whole key (starts with isk_), no missing characters or stray spaces. Can't find it? You can't recover it - delete it on the key page and create a new one.
• The tool doesn't show the server - fully quit and reopen the tool (not just close the window); configs are only read on a fresh start. Cursor/VS Code/Windsurf also have an MCP Refresh button.
• Worked yesterday, not today - a key works only until it's deleted or replaced. If you made a new key, update every tool and scheduled digest with it.
• Is it safe to put the key in a file? - Yes. It only reads jobs, is rate-limited, and you can delete it anytime. That's why we embed it directly instead of fiddling with system environment variables - fewer things to break.

Still stuck? Email support@interviewstack.io with the tool you're using.

---

Step 3 - Daily digest on autopilot

This is the step that makes the whole thing passive: a scheduled run finds what's new for you each morning, auto-saves the strongest matches (each with a one-line "why it fits" note) into your application tracker, and writes a compact digest to a log. You review and apply from the tracker - jobs you remove are never re-saved.

The easiest way: just ask your assistant - "set up my daily job digest to run on a local open model." It will run the installer, set up Ollama, and schedule it. Or do it yourself:

Recommended - free + private, on a local open model

The recurring run uses a local open model (via Ollama) - no hosted-AI key, no per-token bills, and your resume/profile never leaves your machine. After setup, each morning is just python3 digest.py.

# Installs Ollama if needed, pulls the model, and writes a JITTERED schedule.
INTERVIEWSTACK_MCP_KEY=isk_... WINDOW_START=6 WINDOW_END=9 \
  ./examples/cron/open-model/setup.sh

Then edit ~/.config/interviewstack-digest/config.json (your profile + search criteria) and test with python3 ~/.config/interviewstack-digest/digest.py --dry-run. Full guide: examples/cron/open-model/.

Alternative - hosted AI each morning

Prefer to spend a hosted-AI call per run instead of running a local model?

DIGEST_PROMPT="Run my daily job digest: new senior ML engineer roles, remote, US, \
posted in the last 3 days, deduped against my last run. Top 6, save the best 2-3 \
to my tracker with concrete fit reasons." \
WINDOW_START=6 WINDOW_END=9 ./examples/cron/setup-daily-digest.sh

This template runs Claude Code headlessly; on another tool (Codex, Copilot, ...), schedule that tool's equivalent headless command with the same jittered timing.

⏰ Always jitter the schedule. Don't run it at 09:00 sharp. If everyone runs on the hour, the shared job database takes a synchronized hit and gets slow for everyone. Both setup scripts randomize the time for you; if you schedule manually, pick a random minute (e.g. 07:23, not 08:00). The digest isn't time-critical.

---

Starter prompts - what to ask it

Copy-paste any of these into your connected AI tool. They're ordered roughly by where you are in a job search.

Explore what the board covers

"What job search tools do you have from InterviewStack, and what roles does the board support for someone with my background? Here's my resume: [paste]"

Find roles that actually fit you (not keyword soup)

"Here's my resume: [paste]. Find roles on the InterviewStack board that genuinely fit my background - rank by real fit, be honest about stretches, and show salary where listed."

Targeted search with hard constraints

"Find senior backend engineer roles, remote, US, posted this week, paying $180k+. I need visa sponsorship. Top 10, newest first."

Build a shortlist and save it for later

"Search for staff product manager roles in fintech, hybrid or remote in NYC. Save the 3 best fits to my InterviewStack tracker with a concrete reason for each, and give me the link to review them."

Tailor applications to the best matches

"Here's my resume: [paste]. Find the 2 best-fitting senior data scientist roles posted this week and draft tailored resume bullets, a short cover note, and the apply link for each. Save both to my tracker."

Polish your resume against real demand

"Pull 8-10 real senior DevOps postings from the InterviewStack board, tell me which skills and phrasings keep recurring, and rewrite my resume toward that demand. Here's the current version: [paste]"

Scope a career pivot with real data

"I'm a data analyst who wants to move toward ML engineering. Using the InterviewStack board: which adjacent roles exist, what do their postings ask for that I don't have yet, and which current openings would accept my profile today?"

Market recon

"What are the highest-paying remote machine learning roles on the board right now? Which companies show up most in those results?"

Set up the recurring digest (interactive alternative to Step 3)

"Set up a daily job digest for me: new senior frontend roles, remote, EU, every morning at a randomized time between 7 and 9. Each run, save the best 2-3 new matches to my tracker with fit reasons and keep the digest to 6 lines."

A few habits that get better results: paste your real resume (fit ranking is only as good as what the agent knows about you), state hard constraints explicitly (visa, salary floor, location), and ask it to be honest about stretches - the board's curated taxonomy means "no good matches" is a real, useful answer.

---

Use a local, open model (free and private)

Prefer to run the AI on your own machine? Point a local open-source model at the same MCP server. Two real wins:

• Free. No API keys, no per-token bills. The model runs on your hardware; your only cost is your own compute.
• Private. Your model and your resume stay on your machine. (Job search still calls our hosted MCP, so the job data comes from InterviewStack.io, but your resume and the model never leave your computer.)

What you need

1. A local runtime. Ollama is the simplest.
2. A model that is good at tool calling - this workflow is multi-step (the agent calls find_roles, then search_jobs, then get_job), so an agentic/coding model is the right pick. A strong, ~30B example is qwen3-coder:30b (tuned for tool use; budget roughly 24 GB+ of RAM/VRAM at a 4-bit quant). Smaller 7-14B models work but are less reliable at the multi-tool flow. (If you specifically want a ~27B model, Google's gemma3:27b is another option, but pick one trained for tool calling.)
3. An MCP-capable agent client that supports local models, such as Cline for VS Code.

Recipe: Cline + Ollama + Qwen3-Coder

# 1. Pull the model (see ollama.com/library for current tags)
ollama pull qwen3-coder:30b

// 2. In Cline, add the MCP server (MCP Servers -> Configure), remote HTTP:
{
  "mcpServers": {
    "interviewstack-jobs": {
      "type": "streamableHttp",
      "url": "https://mcp-job-search.interviewstack-io.workers.dev/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

1. In Cline, choose Ollama as the API provider and qwen3-coder:30b as the model.
2. Drop AGENTS.md into your workspace so the model follows the curated-filters-first workflow, then ask it to find roles and tailor to your resume.

Honest note: local models are less reliable at multi-step tool orchestration than frontier hosted models, so expect to guide them a bit more. The curated-filters-first design (resolve the role with find_roles, then a structured search_jobs) keeps each step simple, which helps smaller models stay on track. If your client can't reach the remote server directly, bridge it with mcp-remote (see "Any other MCP client" above).

---

What you get

Tools (via the MCP server)

• search_jobs - live search with the board's curated classification (role, level, skills, location, work mode, salary, benefits, company size, …). Search by exact city, whole metro area, or radius / "near me", not just city name.
• get_job - full job detail + the apply link (Pro; a teaser on the free plan).
• save_job - save a well-matched job (with a fit note) to your InterviewStack application tracker - review and apply later at app.interviewstack.io/sidenav/my-applications (Pro).
• list_saved_jobs - read your tracker back from any connected device (save on your laptop, review on your phone): newest first, with each job's status and your saved note. Optional status filter.
• find_roles / find_skills / find_locations / find_companies / list_filter_options
• resolve what you want to the board's canonical taxonomy (more accurate than free-text title matching).

See the full search_jobs filters reference (every filter, its accepted values, and the roles-vs-categories style distinctions), or call list_filter_options / the find_* tools for the live value lists with counts.

Skills / workflow guidance

The same workflows, delivered two ways: as Claude Code skills (auto-installed by the plugin) and as a portable AGENTS.md for every other tool.

• tailor-application - find roles that genuinely fit you, then draft tailored resume bullets + a cover note + the apply link for the best matches. Drafts only; you review and submit. "Find roles that fit my background and help me apply."
• resume-polish - sharpen your resume, optionally targeted at a real posting or role from the board, grounded in what employers are actually asking for. "Make my resume stronger for senior data scientist roles."
• daily-job-digest - a recurring, low-noise roundup of fresh roles matching your criteria, on your own scheduler - it auto-saves the strongest matches to your application tracker so they're waiting for you. "Send me new ML engineer jobs every morning."

On other tools: add the MCP server (Step 2) and drop AGENTS.md in your project - that gives the agent the same curated-filters-first workflow the Claude skills encode. Individual skill files are also in skills/ if you'd rather copy one into your tool's rules.

---

Free vs Pro

Every signed-in account gets a free key. What you get back depends on your plan:

	Free	Pro
Role, level, location, work mode, salary, skills, job type, posting age	Yes	Yes
Company names, full job descriptions, apply links	Hidden (teaser results)	Yes
Filter or look up by company (companies, find_companies)	No	Yes
Save to your application tracker (save_job)	No	Yes
Daily limits	Free limits	Higher Pro limits

On the free plan, results come back marked locked: true with an upgrade link, and the skills and AGENTS.md tell your assistant to say so plainly and never guess the hidden company or apply link. The daily digest still runs as a teaser digest. Upgrade at app.interviewstack.io/user-settings; the next call returns full results, nothing to reconfigure.

Limits & fair use

Each key is read-only, rate-limited, and has a daily cap on requests and on jobs returned (anti-abuse). Free and Pro tiers differ; your current limits show on the key page. If you hit a limit you'll get a clear "try again" message - the skills are built to stay well within the caps.

Privacy & honesty

• The skills never fabricate experience, skills, or metrics - tailoring means reframing what's true, never inventing.
• Your AI drafts; you review and submit. Nothing is auto-applied.
• Job descriptions are treated as data, never as instructions.

Updating

/plugin marketplace update interviewstack-jobs

New skills and improvements ship here - update to get them.

Support

Issues and ideas: open a GitHub issue, or reach us via InterviewStack.io. Jobs are sourced from InterviewStack.io - cite it as the source when you share roles.

License and scope

The contents of this repository (skills, plugin manifests, docs, and example scripts) are released under the MIT License - see LICENSE. Use them freely.

This open license covers the client-side assets only. The InterviewStack.io job-search service, API, and job data are proprietary and operated by InterviewStack.io; access is via an API key and subject to its terms and rate limits. In other words: the client is open, the service is hosted.

Trademark: "InterviewStack" and "InterviewStack.io" are trademarks of InterviewStack.io. The MIT license covers the code, not the marks. Please do not use the name or logo in a way that implies endorsement of a fork or derivative.

Contributing

Issues (bugs, ideas, questions) are welcome. Pull requests are maintainer-led; small fixes are welcome, and for larger changes please open an issue first. See CONTRIBUTING.md and SECURITY.md.