lusha-talent-sourcing
Lusha Talent Sourcing is an AI-powered talent sourcing and candidate data platform for recruiters, sourcers, and talent acquisition teams. I
Open source Open in the app JSON README (API)
About
Lusha Talent Sourcing is an AI-powered talent sourcing and candidate data platform for recruiters, sourcers, and talent acquisition teams. It helps answer two critical recruiting questions: Who is the best-fit talent for this role, and when might the timing be right to reach out? Find and source passive candidates based on job title, skills, level, company, industry, location, and other criteria relevant to your search. Use natural-language candidate search to narrow a broad talent pool to the people who best match your hiring needs. Go beyond candidate discovery with verified contact details, including emails and phone numbers, so you can connect directly with the right talent. Then use candidate mobility and career-move signals, including tenure in the current role, a change in direct manager, and declining headcount at the current company, to understand when someone may be more open to a new opportunity. Use Lusha for talent sourcing, candidate sourcing, passive candidate search, ta
Details
- Kind
- Plugins
- Topic
- Productivity
- Publisher
- lusha-oss
- Origin
- gemini
- Category
- ferramentas
- Version
- 0.2.2
- Last push
- 2026-09-03T11:35:04Z
- Repository state
- ativo
- Language
- JavaScript
- License
- MIT
- Added
- 2026-09-02 19:00:43
- Updated
- 2026-09-03 15:01:09
- Origin id
lusha-oss/lusha-mcp-talent-sourcing-plugin
README
# Lusha for Recruiting — MCP Plugin Source talent for an open position, keep a saved talent pipeline current, and catch the weeks after a leadership change — from inside your AI assistant. Supports **Claude Code** (CLI / Cowork), **Cursor** (plugins), **Codex** (plugins), **VS Code Copilot** (GitHub Copilot Chat with MCP), and **Gemini CLI** (extensions). This is the recruiting counterpart of [lusha-mcp-plugin](https://github.com/lusha-oss/lusha-mcp-plugin). Same shape, different surface: it connects to the Lusha recruiting MCP deployment, which serves 26 tools under recruiter-facing names (`talent_search`, `candidate_profile`, `list_read`) rather than the general Lusha surface. ## Skills | Skill | What it does | |-------|-------------| | `source-movable-talent` | A position to fill → an ordered shortlist of people who fit it and have been in their job long enough to be open to a move, each with a reason to approach them now | | `keep-a-list-live` | A pipeline saved weeks ago → only what changed since the last check: who moved, who was promoted, who newly crossed the tenure threshold, what happened at their employer | | `leadership-change-sourcing` | An executive departure or replacement → the people one level below, grouped by company, each traced to a dated event | All three load the same shared reference: terminology, fifteen hard rules, the tenure mechanism, the ordering rules and the cost table. The canonical copy is `skills/_shared/SHARED-REFERENCE.md`; each skill carries a generated mirror at `references/shared-reference.md`, because hosts differ on whether they bundle a directory outside the skill folder. CI enforces byte-equality. ## How it works Every client loads the **same** `skills/*/SKILL.md` files and the same MCP server. Only the manifest and the store segment differ: | Client | Manifest | MCP endpoint | How to invoke | |--------|----------|--------------|---------------| | Claude Code | `.claude-plugin/plugin.json` | `mcp-hr.lusha.com/mcp/claude` | `/source-movable-talent`, `/keep-a-list-live`, `/leadership-change-sourcing` | | Cursor | `.cursor-plugin/plugin.json` | `mcp-hr.lusha.com/mcp/cursor` | Skills activate from natural language | | Codex | `.codex-plugin/plugin.json` + `mcp.json` | `mcp-hr.lusha.com/mcp/codex` | Skills activate from natural language | | VS Code Copilot | `.github/plugin/plugin.json` | `mcp-hr.lusha.com/mcp/copilot` | `/source-movable-talent`, etc. | | Gemini CLI | `gemini-extension.json` | `mcp-hr.lusha.com/mcp/gemini` | Gemini activates the matching skill on demand | Skills name tools by their bare recruiter-facing name, so one skill source works on every client. Each manifest also sends `X-Lusha-Plugin: <store>-hr` and `X-Lusha-Plugin-Version`, which is how usage from this plugin is attributed separately from the general one. **The host is the whole selector.** The recruiting surface is a dedicated deployment, not a path: `mcp-hr.lusha.com` serves recruiter names on every mount, and `mcp.lusha.com` serves the general ones. There is no `/mcp/hr` path — pointing a manifest at one gets an unknown store. ## Prerequisites - A Lusha account with API access, and the recruiting surface enabled for it ## Install ### Claude Code (CLI / Cowork) ``` /plugin marketplace add lusha-oss/lusha-mcp-talent-sourcing-plugin /plugin install lusha-talent-sourcing ``` ### Cursor Cursor reads `.cursor-plugin/plugin.json` and discovers the bundled `skills/` automatically. Add the repo as a plugin marketplace, then install from `.cursor-plugin/marketplace.json` (catalog `lusha-recruiting-plugins`, plugin `lusha-talent-sourcing`). ### Codex Codex discovers the plugin through the repo catalog at `.agents/plugins/marketplace.json`, which uses a `url` source pinned to a ref. A `url` source is used instead of a local path because Codex rejects a local plugin path that resolves to the repo root ([codex#17066](https://github.com/openai/codex/issues/17066)) and silently drops symlinks during install ([codex#18863](https://github.com/openai/codex/issues/18863)); cloning over `url` keeps `skills/` as real files at the plugin root. ``` codex plugin marketplace add lusha-oss/lusha-mcp-talent-sourcing-plugin codex /plugins ``` Select the catalog, install, then start a new thread so the skills and MCP tools load. ### VS Code Copilot Requires a VS Code version with agent-plugin support and the GitHub Copilot extension. 1. Open the **Command Palette** (`Cmd+Shift+P` / `Ctrl+Shift+P`). 2. Run **Chat: Install Plugin From Source**. 3. Paste the repository name: `lusha-oss/lusha-mcp-talent-sourcing-plugin`. ### Gemini CLI ``` gemini extensions install https://github.com/lusha-oss/lusha-mcp-talent-sourcing-plugin ``` ## Editing this repo **The manifests are generated. Do not hand-edit them.** `plugin.config.json` is the single source for the name, version, description, display fields and per-client endpoint; `scripts/build-manifests.mjs` writes the five manifests, `mcp.json` and the three catalogs from it, and `scripts/check.mjs` fails if any committed file has drifted. This exists because the display name is Marketing's to change and the description is ~1,400 characters: six hand-synced copies is six chances to update five of them. ``` npm run build # regenerate manifests + mirrored references npm run check # the CI gate, four checks ``` ### Testing against a different endpoint To point the plugin at a different recruiting MCP deployment (a staging environment, or a locally-run one), set the `LUSHA_MCP_ORIGIN` env var before building rather than editing `plugin.config.json`: ``` LUSHA_MCP_ORIGIN=https://<your-endpoint> npm run build ``` The committed endpoint is then never the thing you touch by hand, and `npm run check` — which reads `plugin.config.json` and ignores the env override — fails on a tree still carrying a non-production endpoint, so an accidental local build can't be committed. `npm run check` enforces: 1. Every skill has parseable frontmatter whose `name` matches its folder, with a description long enough to route on. 2. **No sales vocabulary** in anything a host loads — no `prospect`, `ICP`, `buying intent`, `decision maker`, `seniority` (the reference says level) or `lead`, and `contact` only inside `contact details` / `contact information` / `make contact`. Code spans and fenced blocks are exempt, because they carry upstream API field names that are not ours to rewrite. A `<!-- vocab-gate:off -->` region opts out explicitly, which the reference needs in order to state the rule at all. 3. The mirrored shared references are byte-identical to the canonical one. 4. The manifests match `plugin.config.json`, Gemini uses `httpUrl`, and every endpoint, attribution header and version lines up. This repo does not maintain its own copy of which tool names the recruiter MCP server serves — that list is server-owned and changes on the server's schedule, so a skill sending the model at a tool this surface doesn't have is caught server-side rather than by a mirror here that could drift. ### Where the words come from The skills and the shared reference are maintained by the Lusha recruiting team as the source of truth for this plugin's terminology, guardrails and behaviour — if a skill file and any other internal note disagree, the skill file wins. ### Placeholders These are display-only fields, owned by Marketing, and all three live in `plugin.config.json`: | Field | Current value | |-------|---------------| | `display.displayName` | `Lusha Talent Sourcing` | | `display.shortDescription` | `Find best-fit talent, get verified contact details, and see who may be open to a move` | | `display.category` | `Productivity + data` — confirm against each host's accepted category list at submission | | `descriptionTemplate` | The store-listing copy. It reaches Codex as `interface.longDescription` and Gemini as the manifest `description`; every other manifest gets the one-liner instead | The long copy is written by Marketing but still passes the vocabulary gate above, so a draft saying "seniority" or "contact data" comes back as `level` and `contact details`. The machine name is `lusha-talent-sourcing`. It was `lusha-hr` in 0.1.0, so anyone who installed that version reinstalls rather than updates. Treat it as fixed from here: renaming it again after install breaks existing installs. ## License MIT. See [LICENSE](LICENSE).