What Trackly CLI does
Trackly CLI gives MCP-compatible agents access to Trackly’s job discovery, company research, application tracking, and controlled application workflows. The underlying service indexes more than 250,000 jobs across over 5,400 companies and more than 40 ATS types.
Job search supports filters for role, region, state or province, internship level, favorites, recency, minimum annual USD salary, sponsorship, work arrangement, and pipeline status. Agents can retrieve individual job records, search companies semantically, list tracked companies, inspect tracker statistics, and find contacts or network information associated with a company.
The server also exposes application-profile and execution controls. Agents can read profile answers and their provenance, discover missing onboarding fields, update scoped answers with optimistic concurrency, and work with user-approved jobs in an Apply queue. Apply execution tools provide progress, recovery, review handoffs, access deferments, immutable waves, and explicit resume approval. Guided application work stops before final submission and can pause for authentication, CAPTCHA, verification, unexpected origins, or unobservable field state.
How it works
Trackly offers two MCP connection modes. The hosted endpoint uses streamable HTTP at https://mcp.usetrackly.app/api/mcp; users authenticate through OAuth, with Google sign-in available during connector setup. Headless clients that cannot complete interactive sign-in can send a Trackly API key as a bearer token.
The local mode runs the installed CLI as an MCP process over stdio. Local connections provide the CLI’s MCP tool set and are intended for clients such as Cursor, Windsurf, and Claude Code. Hosted connections provide 55 tools, while the local distribution provides 58; hosted and local modes share the main search, company, status, and Apply functionality, with some local-only tools.
The CLI can also be used directly from a terminal. Commands cover job searches, job details, company lookup, statistics, preferences, status changes, contacts, briefs, company workspaces, API-key management, and agent setup. Adding --json produces JSON output for CLI commands.
Setup and configuration
For hosted use, add a custom connector in a compatible AI client and enter the Trackly MCP URL, then complete sign-in. Claude Code can register the same endpoint with:
claude mcp add --transport http trackly https://mcp.usetrackly.app/api/mcp
For local use, install Node.js 20.20+ or 22.22+, then install and authenticate the package:
npm install -g trackly-cli
trackly login
trackly agent setup --client both
trackly agent doctor
Claude Code can register the local process with claude mcp add --scope user trackly -- trackly mcp. Cursor and Windsurf can use the CLI’s stdio MCP configuration. An API key can be saved with trackly config --api-key <key> or supplied for an individual command with --api-key; --base-url is available as a one-off global option.
Tools and capabilities
- Search jobs with structured filters or natural-language queries through
trackly_ask.
- Retrieve job details, network briefs, contacts, company workspaces, and company lists.
- Read statistics, discovery preferences, experience limits, application profiles, and onboarding requirements.
- Mark jobs as applied, saved, or dismissed, and maintain role-specific experience limits.
- Request new companies and inspect application queues and execution progress.
- Recover exact Apply candidates, review handoffs, parked members, and access deferments without substituting unconfirmed jobs.
Limitations and notes
Natural-language trackly_ask searches are limited to 20 uses per day. Company requests are limited to five pending requests. Profile updates use optimistic concurrency, and revoking sensitive profile storage requires a two-step confirmation.
Apply automation is deliberately fail-closed. It does not submit applications automatically: the agent fills user-approved forms and stops before Submit for manual review. Greenhouse has full support in the local guided workflow; Ashby and Lever are best effort, while other supported ATS classes use constrained guided mode. LinkedIn-hosted forms and unverified origins remain manual-only. Browser capability, resume data, authentication state, and exact field state are checked during execution.