# Interviewstack Jobs > Official InterviewStack.io job-search tools for AI assistants. Search live curated jobs, polish your resume, and draft tailored applications from your terminal. - **Type:** MCP server - **Install:** `agentstack add mcp-interviewstackio-interviewstack-jobs` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [interviewstackio](https://agentstack.voostack.com/s/interviewstackio) - **Installs:** 0 - **Category:** [Search](https://agentstack.voostack.com/c/search) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [interviewstackio](https://github.com/interviewstackio) - **Source:** https://github.com/interviewstackio/interviewstack-jobs ## Install ```sh agentstack add mcp-interviewstackio-interviewstack-jobs ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # InterviewStack.io Job Search - for your AI assistant Search **live, curated jobs** from the [InterviewStack.io job board](https://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](https://modelcontextprotocol.io) 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`](./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) - sign in at [app.interviewstack.io/sidenav/job-search-mcp](https://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): ```bash 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](#step-2---connect-your-tool). 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](https://app.interviewstack.io/sidenav/my-applications) with a note on why each one fits. You review and apply. Jobs you remove are never re-saved. More ideas: [starter prompts](#starter-prompts---what-to-ask-it). Prefer to script the schedule yourself instead of asking in chat? See [Step 3](#step-3---daily-digest-on-autopilot). --- ## Step 1 - Get your API key (all tools) Sign in at **[app.interviewstack.io/sidenav/job-search-mcp](https://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 `. Claude Code **Most reliable - one line in your Terminal** (key embedded, no env vars): ```bash 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**. Cursor Add to `~/.cursor/mcp.json` (global - works in every project). Paste your real key in place of `isk_your_key_here`: ```json { "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`](./AGENTS.md) into your project root (Cursor reads it), or add it under `.cursor/rules/`. VS Code + GitHub Copilot (agent mode) Add to `.vscode/mcp.json`: ```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`](./AGENTS.md) in the repo root - Copilot reads it (or `.github/copilot-instructions.md`). OpenAI Codex CLI 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`: ```toml 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`](./AGENTS.md) natively - copy it into your project root. Windsurf Add to `~/.codeium/windsurf/mcp_config.json`. Paste your real key in place of `isk_your_key_here`: ```json { "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`](./AGENTS.md) - copy it into your project root. Any other MCP client Point your client at a remote **Streamable-HTTP** MCP server: ```json { "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`](https://www.npmjs.com/package/mcp-remote): `npx mcp-remote https://mcp-job-search.interviewstack-io.workers.dev/mcp --header "Authorization: Bearer YOUR_KEY"`. Then add [`AGENTS.md`](./AGENTS.md) for the workflow guidance. Once connected, grab a prompt from [Starter prompts](#starter-prompts---what-to-ask-it) below. ### Not working? Unblock it Almost every problem is the key. The fastest fix is the **Test connection** button on [the key page](https://app.interviewstack.io/sidenav/job-search-mcp) - 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](mailto: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](https://app.interviewstack.io/sidenav/my-applications), 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](https://ollama.com)) - 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`. ```bash # 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/`](examples/cron/open-model/). ### Alternative - hosted AI each morning Prefer to spend a hosted-AI call per run instead of running a local model? ```bash 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](https://ollama.com) 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](https://cline.bot) for VS Code. **Recipe: Cline + Ollama + Qwen3-Coder** ```bash # 1. Pull the model (see ollama.com/library for current tags) ollama pull qwen3-coder:30b ``` ```json // 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" } } } } ``` 3. In Cline, choose **Ollama** as the API provider and `qwen3-coder:30b` as the model. 4. Drop [`AGENTS.md`](./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 `se … ## Source & license This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [interviewstackio](https://github.com/interviewstackio) - **Source:** [interviewstackio/interviewstack-jobs](https://github.com/interviewstackio/interviewstack-jobs) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/mcp-interviewstackio-interviewstack-jobs - Seller: https://agentstack.voostack.com/s/interviewstackio - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.