lynqu-ai-toolkit

Lynqu AI Toolkit

Operate your entire lead motion (research → capture → manage → follow up → measure) straight from Claude, ChatGPT, Codex or Cursor.

Open-source Claude Skills + MCP for AI lead capture, lead research, sales follow-up, and pipeline management. Powered by Lynqu.

✨ What this does

Connect Lynqu to your AI assistant, then run your sales motion in plain language. No dashboards, no data entry, no context-switching.

You ▸ I met ~30 people at the SaaStr booth. Here's the list [paste].
      Dedupe against our contacts, add the new ones to the Events pipeline,
      tag them SaaStr-2026, and draft follow-ups for the hot ones.

Lynqu AI ▸ Working through it…
   🔎  Checked 30 contacts → 22 new, 8 already in your pipeline (skipped)
   ➕  Created 22 leads in "Events" at the New stage
   🏷️  Tagged SaaStr-2026 · attached to the SaaStr campaign
   🔥  6 flagged hot (asked for a demo). Drafts ready below for your review
   ✅  Done. Want me to send the 6 follow-ups, or tweak them first?

That whole flow is one sentence to you, and a chain of audited Lynqu actions under the hood. The skill is the recipe; the MCP is the hands.

⚡ Quick start

1. Connect the Lynqu MCP server (OAuth, no API key to copy):

Then, in your client:

• Claude (claude.ai, Desktop): Settings → Connectors → Add custom connector, paste the URL, sign in. Claude Code: claude mcp add --transport http lynqu https://api.lynqu.com/mcp/v2
• ChatGPT (Plus, Pro, Business, Enterprise, Edu): turn on Developer mode, open chatgpt.com/plugins → + → Add custom MCP server, paste the URL, keep OAuth, sign in on the Lynqu page. Then call it with @Lynqu in a chat.
• Codex (app or CLI): Settings → MCP servers → Add, Streamable HTTP, paste the URL, then Authenticate. CLI: codex mcp add lynqu --url https://api.lynqu.com/mcp/v2 and codex mcp login lynqu.
• Cursor, VS Code and any other MCP client: add a remote HTTP server with the URL and approve on the Lynqu page.

Full per-client steps, with the gotchas, are in docs/mcp/connect.md. There is also a one-click setup in the Lynqu app under Settings → AI / MCP.

2. Install the skills. The method depends on where you use Claude:

Claude Code (copy the folders)

git clone https://github.com/Gravisun/lynqu-ai-toolkit.git
cp -r lynqu-ai-toolkit/skills/lynqu* ~/.claude/skills/

claude.ai (web, desktop app & mobile) (upload one ZIP per skill)

~/.claude/skills is Claude Code only; everywhere else (including the Claude desktop app) each skill is uploaded as its own ZIP. Needs a Pro, Max, Team or Enterprise plan.

1. Turn on Settings → Capabilities → Code execution and file creation (without it the Skills section does nothing).

2. Grab the ready-made ZIPs from the latest release, or build them yourself. The skill folder must be the root of the archive, not the loose files and not a wrapper directory:

   git clone https://github.com/Gravisun/lynqu-ai-toolkit.git
   cd lynqu-ai-toolkit/skills
   for s in lynqu*; do zip -r "../$s.zip" "$s"; done
   

3. Settings → Customize → Skills → Add and upload the ZIPs one at a time.

⚠️ Writing your own skill? Keep description under 200 characters. Anthropic's documented ceiling is 1024, but the description sits in context permanently and is the text Claude matches your request against. Short and specific triggers better, and uploads cleanly on every surface. python3 scripts/validate_skills.py enforces it.

3. Just ask:

"Research 10 fintech targets in London, qualify them for Lynqu, and create leads for the best fits."

That's it. See examples/prompts.md for a full prompt library.

💡 Plan note: connecting is free on any account. Running tools needs an AI-tier plan (Pro+AI, Business+AI or Enterprise).

🛠️ The skills

Seventeen skills covering the whole motion. You don't need to learn them. Say what's going on and lynqu works out which ones to run, in what order.

/lynqu we just got back from SaaStr with 200 badges and I don't know where anything is

Routes to event → capture → qualify → outreach, confirming before each write.

Or drive them directly:

Every skill ends by writing to Lynqu: a scored lead, a dated task, a note the next rep inherits. A brief nobody acts on is a document; a lead in the right stage with a next step is a pipeline.

🧠 How it works

 You (plain language)
      │
      ▼
 ┌──────────────────────┐     loads the recipe
 │   AI assistant       │ ◄───────────────────────  Lynqu Agent Skill  (this repo)
 │ (Claude / ChatGPT /  │                            • which tools to call
 │  Cursor)             │                            • in what order
 └──────────┬───────────┘                            • the guardrails
            │ calls MCP tools
            ▼
 ┌──────────────────────┐     OAuth 2.1 (no keys)
 │  Lynqu MCP server    │ ◄───────────────────────  https://api.lynqu.com/mcp/v2
 └──────────┬───────────┘
            ▼
 ┌──────────────────────────────────────────────────────────────┐
 │  Your Lynqu workspace                                          │
 │  cards · contacts · leads · campaigns · events · pipeline      │
 └──────────────────────────────────────────────────────────────┘

A typical multi-step ask fans out like this:

"Research fintech targets, create leads, set them up, and draft intros"

  Phase 1 ▸ RESEARCH      lynqu-lead-research   → score & shortlist accounts
  Phase 2 ▸ CREATE        create-lead × N       → leads land in the pipeline
  Phase 3 ▸ ORGANIZE      lynqu-lead-management  → stage, owner, tags
  Phase 4 ▸ OUTREACH      lynqu-sales-followup   → drafts for your approval

You stay in control: skills confirm before writing in bulk and always ask before sending email. Deletes, merges and AI employee approvals each get their own yes, one record at a time.

🔁 Follow-ups, replies and AI employees

Lynqu closes the engagement loop, and the MCP reaches every part of it:

• Personal follow-ups. Every rep can save their own version of a team template or sequence. It replaces the team text on their own leads, unless an admin locked it.
• Follow-up performance. Sent, delivered, opened, clicked, replied, booked and won, by template, sequence step, rep or AI employee. Reply rate is the headline, and nothing is ranked on fewer than 20 delivered emails.
• Replies. With reply tracking on, a human reply stops the lead's automated follow-ups, lands on its timeline and in the outbox, and is forwarded to whoever the email went out as.
• Sequences. Set up a follow-up sequence from the assistant: fire on a stage, a score or a predicate, up to five timed steps, a cooldown, scoped to an event or a segment. Team or personal, and nothing sends until the org's postal address is on file.
• Segments. A named lead filter (market, source, campaign, tags, UTM, custom fields) that works as a list filter, a sequence audience and a report row. The segment report says which market converts, and the rows add up.
• AI employees. Read what an agent did and why, see the work it handed back, and approve or reject its parked actions one at a time. Agents write from the org's confirmed sales brief.

How these fit together: docs/concepts.md.

💬 Examples

Lead research → pipeline

"Build a list of 10 mid-market SaaS companies in DACH attending SaaStr Europe, find a VP Sales or RevOps lead at each, score them for Lynqu, and create leads for anything 7+."

🔎 Researched 14 → shortlisted 10 · scored on ICP fit + buying signal
   ⭐ 9/10  Northwind Cloud: hiring 4 AEs, VP Sales identified
   ⭐ 8/10  Veridian Pay: Series B last month, RevOps lead identified
   …
➕ Created 7 leads (score ≥ 7) · 3 held for your review (4–6)

Post-event follow-up