Connect your AI client

TekJobs exposes its tools through a local MCP server, tekjobs mcp, speaking JSON-RPC over stdio. Every client below runs that command and talks to it. None of them needs a separate model API key: TekJobs adds no per-call AI billing, and usage follows the client you already have. The server finds your profile folder the same way the CLI does: TEKJOBS_PROFILE if set, else ~/.tekjobs/config.json (which tekjobs init wrote), so it does not matter which folder your client is opened in.

Install the command first, and check it answers:

npm install -g @timurtekb/tekjobs
tekjobs --version

Claude Code

claude mcp add tekjobs -- tekjobs mcp

Then, in any Claude Code session:

Use the tekjobs MCP server. Call onboarding_status, then onboarding_materials, and follow its script: interview me, write my profile, set the criteria, run a dry scan, and show me the top matches.

From a clone of the repository you do not need the add: app/.mcp.json connects the server whenever Claude Code is opened in app/.

On Windows, if the server shows as failed to start, the npm shim needs a shell: claude mcp add tekjobs -- cmd /c tekjobs mcp.

Codex

codex mcp add tekjobs -- tekjobs mcp
codex mcp list

Or by hand, in ~/.codex/config.toml:

[mcp_servers.tekjobs]
command = "tekjobs"
args = ["mcp"]

Then open Codex and use the same onboarding prompt as above.

Cursor and Claude Desktop

Both take a JSON block in their MCP settings: Cursor under Settings, MCP (or .cursor/mcp.json in a project); Claude Desktop in claude_desktop_config.json (Settings, Developer, Edit config).

{
  "mcpServers": {
    "tekjobs": {
      "command": "tekjobs",
      "args": ["mcp"]
    }
  }
}

Restart the application after saving. On Windows, use "command": "cmd", "args": ["/c", "tekjobs", "mcp"] if the server does not start.

What about ChatGPT?

ChatGPT's web app does not connect to a local stdio MCP server. Its developer mode reaches remote MCP servers only, on eligible plans, which would mean putting TekJobs behind a tunnel with authentication in front of it. That is not the standard setup and not one we document. For an OpenAI model, use Codex with the local server above; it is the same tools and the same notes.

What each client can and cannot do

Every client above can run the interview, search and move jobs, draft letters and resumes into the notes, add boards, and start scans: the full tool list is on the MCP server page. Two things are client-specific:

• Check mail needs Claude Code, with its Gmail connector enabled. The run allows three Gmail read tools by name and denies every write and shell tool, which is what makes it safe to run unattended. Another client would need a Gmail MCP with matching tool names. Everything else works without it.
• The app's own Write cover letter and Tailor resume buttons run a command-line LLM with the prompt as an argument. The default is Claude Code's claude -p; another command goes in Settings. Over MCP, any client drafts the same material through cover_letter_materials and tailored_resume_materials.

Troubleshooting

The client cannot find tekjobs. Confirm the global install answered (tekjobs --version), then restart the client; most read their MCP configuration at start. If you installed with npx only, the command exists in npx's cache and not on your PATH: run npm install -g @timurtekb/tekjobs.

The server appears, but with no tools. Run tekjobs mcp in a terminal. A healthy server waits silently for JSON on stdin; an error about the profile folder means tekjobs init has not run, or ~/.tekjobs/config.json points somewhere that has moved.

The interview cannot read my resume. tekjobs status says what the onboarding still needs, and tekjobs resume <file> imports or replaces the resume (PDF, DOCX, Markdown, text).

I want to try it without my own data. The repository ships a fictional profile folder, samples/vault; point TEKJOBS_PROFILE at it and the same client sees Jordan Example's search instead of yours. See A look before you commit.