Email intelligence for every role in the company. A Claude Code plugin marketplace that turns the email already in your inbox into structured, queryable insight — by role, by skill, in plain language.
13 role-specific plugins · 76 focused skills · 13 agents · all powered by the iGPT MCP at .
Email intelligence for every role in the company, 13 role-specific plugins, 76 skills, and 13 agents for Claude Cowork and any MCP-compatible agent. Powered by the iGPT MCP at https://mcp.igpt.ai/.
Sterne
17
Tools
—
Uptime
—
https://mcp.igpt.ai/
What this is
iGPT Skills is a marketplace of Claude Code plugins that lets people ask plain-language questions about their email and get back grounded, structured answers — by role.
Each plugin targets a specific role (sales, finance, HR, ...) and ships:
A tightly-scoped set of skills that each answer one focused question.
An agent that handles broad, cross-skill questions and synthesizes a prioritized response.
An MCP configuration that points at the iGPT backend at https://mcp.igpt.ai/.
A salesperson asking "who's waiting on a reply from me?", a finance lead asking "build me an expense report for May in USD", a researcher asking "what grant deadlines are coming up?", or a CSM asking "which customers should I be worried about?" — the right skill picks itself up and returns a clean, ready-to-act answer grounded in their actual email threads.
Who it's for
End users. Salespeople, customer success managers, finance leads, HR business partners, recruiters, executives, marketers, real estate agents, consultants, procurement leads, IT teams, researchers — anyone whose work runs through email.
Builders and integrators. Developers extending the plugins, customizing prompts and schemas, or integrating the iGPT API directly. Every skill's prompt and structured output schema is documented at the file level (plugins/<plugin>/skills/<skill>/SKILL.md) and reusable.
Quick start
Install one of the plugins and connect your email through iGPT. That's the whole setup.
Cowork (Claude desktop app)
Open the plugins section in Cowork. Search the marketplace for iGPT, or add from GitHub: paste igptai/skills (or the full URL https://github.com/igptai/skills).
Pick the plugin matching your role (e.g. igpt-sales).
Install. The first time you ask the plugin a question, you'll be prompted to sign in to iGPT (free account; OAuth in a browser; read-only email permissions).
Ask in plain language: "who's waiting on a reply?", "what grant deadlines are coming up?", "any compliance flags I should know about?"
Claude Desktop
Settings → Connectors → Add MCP Server → enter https://mcp.igpt.ai/ and complete OAuth.
Claude Code
/plugin marketplace add igptai/skills
/plugin install igpt-sales@igpt-skills # or any plugin name
The plugin ships its own .mcp.json so the MCP server is registered for you. OAuth happens on first use.
Other MCP clients
Any Streamable-HTTP MCP client works. Point it at https://mcp.igpt.ai/ and use the prompts and JSON schemas from each SKILL.md directly.
Every plugin has a dedicated README.md (linked above) with chat-style examples, a "How it works" pipeline, and hand-held install steps tuned to the role. The skill tables below give a one-line summary of what each skill solves.
igpt-sales — Sales intelligence
For account executives, sales leaders, and BDRs working active deals.
Skill
What it solves
deal-open-items
Every unresolved commitment, decision, and "I'll get back to you" across active deal threads
deal-friction-detector
Objections, hesitations, pricing friction, and stall signals on a specific deal
cross-sell-upsell-signals
Adjacent problems, budget hints, or expansion interest buried in existing customer email
competitor-mentions
Every mention of a competitor or alternative across accounts, with sentiment context
follow-up-radar
Threads waiting past silence thresholds — them on us, us on them, unanswered questions
meeting-prep-briefing
Pre-call brief on a contact at a company — history, open items, concerns, what they care about
past-solutions-matcher
Past customers who faced a similar problem — fuel for case-study moments in conversation
igpt-cs — Customer success intelligence
For CSMs, account managers, and customer success leaders.
Skill
What it solves
churn-signal-detector
Early warning signals across accounts — dissatisfaction, competitor mentions, declining engagement
renewal-readiness-checker
Renewal sentiment, engagement, open issues, likely-to-renew vs. at-risk
escalation-tracker
Active escalations, SLA breaches, executive complaints, unresolved issues with growing severity
Papers, preprints, datasets shared in email — relevance to ongoing projects, follow-up actions expected
research-action-items
Action items assigned to lab members — analyses, experiments, code, writing, reviews
Architecture
Each plugin in this marketplace bundles four kinds of artifact, all auto-discovered by Claude Code from convention-based folder names.
Skills
A skill is a single focused workflow: collect a few inputs from the user, query iGPT against their email, return a structured answer.
Each skill lives at plugins/<plugin>/skills/<skill-name>/SKILL.md and follows a consistent template:
Step 1 — collect user-supplied values. Variables like [time_range] and [scope] are defined with defaults and descriptions. The agent or skill asks the user for any value without a default.
Step 2 — query iGPT with structured parameters (search-style filtering by date range, keywords, scope).
Step 3 — call the iGPT ask tool with the user's variables substituted into a prompt template, and a strict JSON output schema.
Skills always return schema-validated JSON to the LLM, which is what produces the clean rendered chat response the user sees.
Agents
An agent is a higher-level coordinator for a plugin. Each plugin ships exactly one agent at plugins/<plugin>/agents/<plugin>.md that handles broad, cross-skill questions ("walk me through what needs my attention this week") by routing to the right skills, running them in parallel, and synthesizing a prioritized response.
Agents also enforce domain-specific guardrails:
Verbatim commercial terms in finance and procurement
Patterns-not-judgments in HR and recruiting
Stage discipline + contrarian signals in sales
Faithful decision reconstruction in projects
And so on across all 13
Plugins
A plugin is a folder containing one set of skills, one agent, an .mcp.json for MCP server registration, and a plugin.json with metadata. Plugins are auto-discovered by Claude Code from the marketplace manifest.
Marketplace
The marketplace manifest at .claude-plugin/marketplace.json lists all 13 plugins with their categories, keywords, and source paths.
How it works
You ask in chat
↓
Agent (or skill) routes the question
↓
Skill queries iGPT against your connected email
↓
iGPT returns structured findings (with evidence)
↓
Claude renders the findings into readable markdown
A plain-language question becomes the right query against your email, returns evidence-grounded structured findings, and gets rendered into a readable answer. You see the rendered answer in chat; integrators can also work with the structured output directly.
Privacy and data flow
Your email stays with iGPT. The plugin doesn't pipe raw email to Claude. You connect your email account to iGPT through OAuth (read-only); iGPT processes the email on its side; only the structured findings — never the raw inbox — reach Claude.
OAuth, not credentials. No API keys, passwords, or access tokens are ever shared with the plugin or pasted into chat. Authentication happens through a standard OAuth flow in the browser.
Read-only. The iGPT MCP only reads from your connected email. The skills are designed for analysis and synthesis — they never send email on your behalf.
Per-user authorization. Each user installs and authorizes their own iGPT connection. Multi-user deployments preserve user-level access controls — no cross-user email leakage.
Sensitive content care. Several plugins enforce additional rules at the skill and agent layer:
igpt-recruiting: candidate names handled as sensitive data; hiring manager performance described as patterns, not character judgments.
igpt-executive: measured language for board / fundraising / settlement / compensation contexts; minimized raw quotation.
See each plugin's README.md for the full set of role-specific guardrails.
.mcp.json — MCP server registration (type: "http", URL points to iGPT)
README.md — buyer-facing front door with chat examples, install instructions, output schemas
agents/<plugin>.md — the per-plugin agent
skills/<skill>/SKILL.md — one folder per skill, each with the [var]-driven workflow
Customization and extending
Adjust an existing skill
Each SKILL.md is human-readable and self-contained. Edit the workflow's variable defaults, search keywords, prompt template, or output schema in place. The skill is ready to use as soon as the file is saved — no build step.
Add a new skill
Create a folder under plugins/<plugin>/skills/ with a SKILL.md that follows the project pattern:
YAML frontmatter with name, description, metadata.version
A 3-step workflow: collect user inputs (Step 1), query iGPT (Step 2), call ask with input and output_format (Step 3)
An inline ## Prerequisites block (the standard one used across the marketplace)
Then add the skill to the corresponding agent's "Available skills" list so the agent knows about it.
Add a new plugin
Copy the structure of an existing plugin (e.g. plugins/igpt-consulting/), update plugin.json, write skills and one agent, and add an entry to .claude-plugin/marketplace.json.
Use the prompts and schemas outside MCP
Every SKILL.md documents its prompt template and output schema in plain text. They work directly against the iGPT API without going through MCP — useful for backend integrations, batch processing, or custom UIs. See docs.igpt.ai for direct API docs.
Conventions
This repository follows a small set of strict conventions documented in CLAUDE.md:
.mcp.json shape — {type: "http", url: "https://mcp.igpt.ai/"}, no extra fields
Inline Prerequisites blocks — self-contained per skill, with the canonical shared/mcp-guard.md as fallback
Workflow template — [var] placeholders, step-1 user input collection, step-3 ask with input only
Per-plugin README pattern — practitioner-first voice, "What you can ask", "What you see", "How it works", install paths, collapsible JSON for developers
Per-plugin agent — one per plugin at agents/<plugin>.md with domain-specific guardrails
Contributors should read CLAUDE.md before making structural changes.
Versioning
Each SKILL.md and agent file carries a metadata.version field (currently 1.0.0 across all 76 skills and 13 agents). Bump the version when behavior changes meaningfully — output schema changes, prompt rewrites, default value shifts.
The marketplace itself follows semantic versioning at the plugin level (see each plugin.json).
Contributing
Contributions are welcome. Please read CLAUDE.md first — it documents the project conventions for skills, agents, READMEs, and .mcp.json configs. PRs that follow those conventions land faster.
A few specific notes:
Don't hardcode plugin counts in plugin-level READMEs — say "all our role-specific plugins" so they don't go stale.
Use the [var] placeholder convention in SKILL.md workflows. Hardcoded values inside ask prompts are a no-go.
Domain-specific guardrails live at the agent body level. Surface them in plugin READMEs' "How it works" sections too.
Related
iGPT website — sign up for the iGPT account that backs these plugins
iGPT API docs — direct API reference for backend integrations
Playground — try queries interactively before writing code
License
MIT
S
Skills
igptai·17 Sterne·
Email intelligence for every role in the company, 13 role-specific plugins, 76 skills, and 13 agents for Claude Cowork and any MCP-compatible agent. Powered by the iGPT MCP at https://mcp.igpt.ai/.
Noch keine Tool-Liste. Remote-Server erhalten eine nach dem ersten Health-Check.
FAQ
Fragen
Wie installiere ich Skills?
Kopieren Sie auf dieser Seite die stdio-Konfiguration für Cursor, Claude Desktop, Claude Code oder VS Code. Sie nutzt die Installation des Herausgebers, keinen PlugMCP-Proxy.
Benötigt Skills einen API-Schlüssel?
In diesem Eintrag ist keine erforderliche Umgebungsvariable aufgeführt.
Welchen Transport verwendet Skills?
Skills läuft lokal über stdio.
Wann wurde Skills zuletzt geprüft?
Zuletzt geprüft: 2026-10-27. Email intelligence for every role in the company, 13 role-specific plugins, 76 skills, and 13 agents for Claude Cowork and any MCP-compatible agent. Powered by the iGPT MCP at https://mcp.igpt.ai/.