instaffo-mcp-server
Enables AI assistants to interact with your personal Instaffo candidate account, allowing reading of profile, job suggestions, and conversations, as well as performing reversible write actions like bookmarking jobs and updating skills, all authenticated via your browser session and running fully locally.
README
instaffo-mcp-server
An MCP server that gives an AI assistant access to your own Instaffo candidate account, authenticated with your own browser session. Personal job-search tooling. Runs fully local.
Instaffo is a two-sided recruiting marketplace (candidates are matched with companies). This server exposes the candidate side: read your profile and job suggestions, read conversations, and perform reversible write actions, from an MCP client.
How it works
Instaffo has no public candidate API, so the server authenticates with a real
logged-in session. Its candidate web app talks to a clean JSON API under
app.instaffo.com/candidate/api/v1/*, authenticated purely by the session
cookie (no bearer token). So the server is a thin cookie-authenticated httpx
client, not a scraper. A browser (via patchright) is used only once, to mint
the session at login.
MCP client ── stdio ──> instaffo-mcp-server ──cookie──> app.instaffo.com JSON API
│
storage-state.json (cookies, written 0600, git-ignored)
▲
instaffo-mcp --login (one-time browser sign-in)
Setup
uv sync
uv run patchright install chromium # one-time, for login only
uv run instaffo-mcp --login # opens a browser; sign in once
uv run instaffo-mcp --auth-status # confirm the session is stored
Register it with your MCP client (stdio):
{
"command": "uv",
"args": ["run", "--directory", "/path/to/instaffo-mcp-server", "instaffo-mcp"]
}
Tools
Reads (no side effects):
| Tool | What it returns |
|---|---|
instaffo_whoami |
your identity (name, email, job title) |
instaffo_get_profile |
your profile: seniority, location, skills, CV summary |
instaffo_list_job_suggestions |
your current matches, with counters |
instaffo_get_job_suggestion |
one role in full: description, requirements, salary, screening questions |
instaffo_list_conversations |
company requests (inbound interest) and your applications |
instaffo_auth_status |
is a session present (--deep validates it live) |
Writes (every write tool takes confirm; without it you get a preview and
nothing changes):
| Tool | Effect | Endpoint |
|---|---|---|
instaffo_save_job |
bookmark a suggestion (reversible) | POST .../job_suggestions/{uuid}/favorite |
instaffo_unsave_job |
remove a bookmark | DELETE .../job_suggestions/{uuid}/favorite |
instaffo_set_skill_experience |
set your years per skill on your profile | POST .../experience_durations/bulk_save |
All endpoints above are verified against the live API.
Supervised, on purpose: apply and message
Two actions are intentionally not implemented as fire-and-forget tools: applying to a job, and messaging a recruiter.
Applying is not one request. It is a multi-step wizard that writes lasting self-representations to your real profile before it submits:
- a skill self-assessment (year sliders per required skill, e.g. 0-5), which
auto-saves to your profile via
experience_durations/bulk_save, - your salary expectation (pre-filled from your profile),
- an "AI tools you use" and "AI skills" multi-select,
- a final submit that creates the application and opens a chat with the recruiter.
Because those are real, outward-facing choices about how you present yourself, and the final submit endpoint only appears once the whole flow is completed, the apply and message tools are left for a supervised session where the account owner approves the inputs. They are not built against a guessed endpoint. The observed sub-steps and wizard shape are recorded here so that session is quick.
Commands
instaffo-mcp # run the MCP server (stdio)
instaffo-mcp --login # headed manual login, persist the session
instaffo-mcp --capture # record app API traffic to a JSONL (diagnostics)
instaffo-mcp --auth-status [--deep]
Security and privacy
- Session material (
profile/,storage-state.json,captures/,.env) lives under~/.instaffo-mcp, is written0600, and is git-ignored. It is never committed. - Write tools are confirm-gated and only touch reversible surfaces.
- This is a personal, local tool for your own account. It stores no one else's data and talks only to Instaffo with your own session.
Prior art
The session-capture and browser patterns are adapted in spirit from stickerdaniel/linkedin-mcp-server (Apache-2.0), which authenticates a personal LinkedIn session the same way.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。