This page documents the tool connection layer: the pull layer that lets your coding agent use the tools you already use. For the broader personal AI assistant for work stack, start with the top-level README.
Tool connections fetch context and take action when a workflow asks for it. They complement triggers/, which listen for app and service events and wake workflows up.
Most "AI integration" approaches ask you to:
- Set up cloud middleware (Zapier, MCP servers, hosted agents)
- Request IT-approved service accounts
- Wait for admin provisioning
- Accept new attack surfaces and vendor lock-in
10xProductivity flips this completely.
Apps are built for humans on laptops. Browsers, CLIs, REST APIs — they're all already there. Your agent uses the same surface. No company-wide integration layer, no middleware, no new permissions.
1. Local agent as the universal client Your laptop is the platform. The agent is just a smarter version of you running scripts. No new infrastructure required.
2. Security by locality The threat model is identical to you doing it manually. Nothing new is exposed. No cloud service sits between you and your tools holding your credentials. The only trust you extend is to the agent runtime itself (Cursor, Claude Code, Codex, Copilot, etc.) — which you've already decided to trust.
2b. Agent automation, not mass scraping
Connections target daily agent-driven use — read context, draft, act with approval — not bulk harvesting or high-frequency unattended scraping. When evaluating a connector, prioritize whether it works, whether it exfiltrates data, and write-approval gates; do not treat generic bot-ban warnings (meant for scrapers at scale) as the default reason to reject a viable path. See add-new-tool.md → Operating context.
3. Identity = accountability Your personal token. Your name on every action. This is a stronger audit trail than most enterprise automation, where actions are taken by service accounts with shared credentials. The agent acts as you: you get the credit, you get the blame, and the audit log is already there in every system you use.
4. Zero friction to start
No OAuth app approval. No IT ticket. No staging environment. If you can log in, your agent can act. Clone the repo, point your agent at setup.md, and it handles the rest.
Most AI tools — enterprise search platforms, knowledge bases, even ChatGPT — are read-only. They find information and surface it in a chat box. You then manually take that answer and do something with it.
10xProductivity is read + write + act. The agent doesn't just find the Jira ticket — it can update it when the workflow calls for that. It doesn't just summarize the Slack thread — it can post the summary back. It doesn't just locate the bug — it can open the PR that fixes it.
The output isn't text in a box. The output is the thing done, in the right place, in the right tool.
This is a categorical difference. Any AI tool that only answers can be replaced by a better search engine. An agent that acts is a different class of capability entirely.
If every individual is 10x productive, the team and company is 10x as a result. Not through a top-down platform rollout — through individuals who are dramatically more capable, spreading organically.
Each connection you add isn't just a new tool — it compounds with everything else.
Once your agent can see across your tools simultaneously, a new class of capability emerges:
Connected knowledge Ask questions that span systems. "What was the decision behind this change?" pulls the GitHub PR, the Jira ticket it closed, the Slack thread where it was debated, and the Confluence doc that captured the outcome — in one answer, in seconds.
Cross-tool reasoning "Show me all incident alerts from last week that still have open ticket follow-ups with no activity in 3 days." That query touches multiple systems, requires no new integration layer, and runs right now. The agent is the integration layer.
Compound automation Repetitive multi-step work — triage a Slack alert, file a Jira ticket, assign it, post a summary back to the channel — becomes a single agent instruction. The tools are already connected; the only thing left is telling it what to do.
Institutional memory Your agent stops being a generic assistant and starts being a contextual one — aware of how your team works, who owns what, what's in flight, and what happened before. That context doesn't live in any one tool. It lives in the connections between them.
The pre-built recipes in this repo are a starting point. The actual ceiling is "everything you can do on your laptop" — which is everything.
Once your tools are connected, the built-in enterprise search workflow lets your agent query all of them simultaneously — Slack, Confluence, Jira, Linear, Notion, GitHub, and more — and synthesize a single answer.
One prompt, every connected source:
Search for everything related to the decision to deprecate the v1 API.
The agent fans out across every connected tool, pulls relevant results, and returns a synthesized answer with citations — no tab switching, no copy-paste, no hunting across systems.
What it replaces:
- Opening Slack, searching, scrolling, opening threads
- Switching to Confluence, searching again
- Checking Jira, checking GitHub PRs, checking Linear
- Mentally stitching together five partial answers
When to use it: Any time you're asking a question that might be answered in more than one place — "what was the decision on X", "who owns Y", "is there a doc on Z", "any context on this incident", "what did we decide about this feature".
To activate: Read /path/to/10xProductivity/workflows/enterprise-search/enterprise-search.md
Agent-readable playbooks for connecting your local agent to the tools you already use — with no limit on what those tools can be.
tool_connections/ ← pre-built recipes for common tools (e.g. Slack, GitHub, Jira)
slack/
setup.md ← how to connect
connection-sso.md
personal/ ← your own recipes (gitignored) — internal tools stay private, never committed
{tool-name}/
setup.md
connection-{auth-method}.md
workflows/ ← pre-built workflows that compose multiple tool connections
enterprise-search/
enterprise-search.md ← search across all your connected tools simultaneously in one query
add-new-tool.md ← connect anything not in tool_connections/ — internal portals, custom systems, any tool with an API or browser interface
setup-python.md ← detect OS, try winget/brew/apt then python.org if needed; Python 3.12 + venv + Playwright (no git — you already have this folder)
tool_connections/ has pre-built recipes for tools most teams share. personal/ is where your internal company tools live — same setup path, stays on your machine. Once connected, workflows like the built-in search let your agent query Slack, Confluence, Jira, and more in a single request.
The pre-built list is just a head start. add-new-tool.md is a playbook for connecting any tool that isn't there yet — internal portals, proprietary systems, niche SaaS, anything. If it has an API or a browser interface, your agent can use it.
- Install Cursor (the editor your agent runs in): open cursor.com/download, download the installer for Windows or macOS, run it, and sign in or create an account when prompted.
git clone https://github.com/ZhixiangLuo/10xProductivity.git
cd 10xProductivityOpen the 10xProductivity folder in Cursor (File → Open Folder…).
If you do not have Python 3 yet (common on a fresh laptop), point your agent at setup-python.md first — it detects your system, tries winget / Homebrew / apt when present, otherwise guides a python.org install, then prepares .venv + Playwright. No git in that playbook — you already have this folder.
Then point your agent at the setup guide:
Read /path/to/10xProductivity/setup.md and set up my tool connections.
Your agent handles the rest — it will ask which tools you use and the URLs, get the credentials it needs, run SSO where required, and verify each connection works. Works for any tool: pre-built recipes for common tools, and an identical setup path for internal or custom tools.
User — you want to connect your agent to tools you already use:
- Clone the repo and point your agent at it: "Read /path/to/10xProductivity/setup.md and set up my tool connections"
- Your agent asks which tools you use, handles credentials, runs SSO where needed, and verifies each connection
- From that point your tools and the search workflow are available automatically at the start of every session — no MCP server, no plugin, no admin approval
Contributor — you want to add a new tool or improve an existing connection:
- Ask your agent: "Load add-new-tool.md and add a connection for [Tool]"
- The skill walks through: research auth → ask URL first → try the most likely auth → ask only for missing credentials → validate → write → PR (contribution is optional and only for commercial tools)
- Community files (
staging/) have a lower bar; core (tool_connections/) requires multi-environment validation
Contributions are welcome for:
- New tool — a tool not yet in the repo
- New auth variant — a different auth method for an existing tool (e.g. AD SSO vs API token)
- New deployment variant — e.g. Jira Server vs Jira Cloud
- Improvement to an existing connection — fixing broken snippets, adding missing endpoints, updating stale auth
If something doesn't work or you want to request a new tool, open an issue.
See ../contributing.md for the full process. The core rule: run before you write. Every snippet must be code you actually executed and saw succeed. No copy-paste from docs.