Pickle MCP Server

Pickle MCP Server

Free AI ops manager that runs locally, auditing ClickUp, Slack, and Teams to catch stale tasks, dropped promises, and buried decisions using your own tokens.

Category
Visit Server

README

🥒 Pickle: the free AI ops manager for ClickUp, Slack & Teams

Pickle: catch what slips through ClickUp, Slack and Teams. The AI ops manager that catches what your team dropped

ClickUp Brain reads your tasks. Slack AI reads your messages. Both are paid add-ons, and each only sees its own app. Pickle is a free AI ops manager that runs on your own machine, reads across all three, and tells you what slipped through: stale tasks, dropped promises, decisions buried in a DM.

License: MIT MCP compatible No account · no telemetry Works with Star this repo

Pickle is a free, open-source AI ops manager for your team, packaged as an MCP server you run yourself. Give it your own ClickUp / Slack / Teams token, and your AI client (Claude, Cursor, Codex, Cline, Zed…) audits your workspace the way a sharp ops manager would: catching the work that quietly fell through.

No account. No hosted server. No key from us. No telemetry. Pickle adds no middleman. It talks only to your tools, with your token. Nothing is sent to us, because there's no "us" to send it to.

⭐ Pickle is free and will stay free. There is no paid tier. If you think workspace tools should run on your own machine, star the repo. Stars are the only way it gets found.


<a id="privacy"></a>Why this is different

ClickUp Brain Slack AI Pickle
Sees across your tools ❌ ClickUp only ❌ Slack only ✅ ClickUp + Slack + Teams
Where your data flows Their cloud Their cloud Only to your own tools, with your token
A third party holds your token Yes Yes No, it stays in your local config
Cost Paid add-on ~$10/user/mo Free · MIT · open source
You can read every line ❌ ❌ ✅ it's all right here

Honest version of the privacy claim: Pickle does read your ClickUp/Slack/Teams. That's the job. But it reads them directly, from your machine, with your token, and sends nothing to any Pickle server. Verify it yourself. The code is in server-remote/server.mjs, and there is no analytics, no beacon, no phone-home.


<a id="install"></a>Install: one command

Guided: it asks which tools you use, points you to where each token lives, and writes your MCP config for you.

curl -fsSL https://pickle.adityaarsharma.com/install.sh | bash

Or add it to your MCP client yourself (via npm):

{
  "mcpServers": {
    "pickle": {
      "command": "npx",
      "args": ["-y", "pickle-mcp"],
      "env": { "CLICKUP_API_KEY": "pk_your_own_token" }
    }
  }
}

Add SLACK_TOKEN (xoxp-…) and/or TEAMS_TOKEN to the same env block to switch on the cross-tool patterns. Restart your client. npx runs it locally on your machine, no global install, no clone.

The installer asks which tools you use (ClickUp / Slack / Teams), takes the token you already have for each, and writes the MCP config for your client automatically. No hand-editing, no key from us.

Prefer to do it by hand? See docs/manual-install.md. The only token you ever provide is your own platform token (e.g. ClickUp pk_… from Settings → Apps).

Then just ask your AI:

"Pickle, run my morning audit: scan my ClickUp from the last 7 days and show me the 3 worst things I missed, worst first."


What it catches

  • Stale in-progress: "doing" for days, zero activity
  • Dropped promises: "I'll do X by Friday" with no follow-through
  • Decisions buried in DMs: a call made in a thread that never became a task
  • Delegations nobody chased: you asked, it never landed, everyone forgot
  • Standup copy-paste, zombie tasks, empty-description work, blocker rot… full list in docs/patterns.md

<a id="works-with"></a>Works with

Claude Code · Claude Desktop · Cursor · Codex · Cline · Continue · Zed · any MCP host. The reasoning runs in your model. Pickle just fetches and structures the data. Optional Claude Code skills add /pickle-* slash commands as a convenience layer.


Free, and what I ask instead of money

Pickle has no paid tier and never will. Instead:

  • ⭐ Star the repo if it caught even one thing worth fixing. A free tool has no ad budget, so a star is genuinely how the next person finds it.
  • 📮 Get update notes + workspace-AI tips: occasional, practical notes on using AI in real team workflows (the same patterns Pickle runs). Reply anytime with a suggestion or a pattern you want added. I read every one.

🧭 Want your team to actually use AI well?

Pickle is what I use to keep my own team honest. If you're rolling AI into how your company works (onboarding people onto it, wiring it into real workflows, choosing which LLM for what), reach out to me. I help teams go from "we bought AI licenses" to "AI is actually in our workflow," with real use-cases instead of hype.

Built in the open by Aditya Sharma, a marketer who codes.

FAQ

Is Pickle really free? Yes. MIT-licensed, no paid tier, no account, no credit card. I built it in the open; a ⭐ is the only thing I ask.

Do I need a Pickle account or an API key from you? No. Pickle issues nothing. The only token you provide is your own ClickUp / Slack / Teams token, and it lives in your local MCP config.

Does my data get sent to you or to any cloud? No middleman. Pickle runs on your machine and talks only to your tools (ClickUp / Slack / Microsoft Graph) using your token. There is no Pickle server, no telemetry, no phone-home. Pickle does read your ClickUp/Slack/Teams (that's the job), but it sends nothing to us. Verify it in server-remote/server.mjs.

Which AI clients does it work with? Any MCP-compatible host: Claude Code, Claude Desktop, Cursor, Codex, Cline, Continue, Zed, and more. The reasoning runs in your model.

What should I ask Pickle first? After installing, restart your AI client and paste: "Pickle, run my morning audit: scan my ClickUp from the last 7 days and show me the 3 worst things I missed, worst first." You get a ranked, punchy result instead of a data dump, so you feel the value on the first run.

How is this different from ClickUp Brain or Slack AI? Those are cloud add-ons that each see only their own tool and send your data to their servers. Pickle runs locally, reads across ClickUp + Slack + Teams, and keeps your token on your machine. Different axis: privacy + cross-tool, not "more features."

Is it safe to give it my token? The token stays in your local config file (chmod 600), and the code is open. Read it before you connect. Nothing is transmitted anywhere except the platform's own API.

Will Pickle change or delete anything in my tools? By default it only reads and reports, it does not touch your data. It can create a reminder or drop a comment if you explicitly ask it to (for example "remind me about XYZ-184"), but the audit itself is read-only. You stay in control.

Do I need Slack and Teams too, or is ClickUp enough? ClickUp alone is a great start. Slack/Teams unlock the cross-tool patterns (decisions-in-DMs, ghost mode) because those need chat data.

Can it audit GitHub or other tools? Not yet. A GitHub audit (stale PRs, dropped reviews) is a planned/welcome contribution.

Is there a hosted version? No. Pickle is local by design. If you want your team to adopt AI tools like this well, reach out. That's the consulting I do.

Contributing

Issues and PRs welcome: new patterns, new connectors (a GitHub audit for stale PRs is a great first one), better client configs.

License

MIT © Aditya Sharma

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured