GTM-engineering walkthrough · Folk MCP · 2026
Folk + Claude via community MCP — the relationship-led CRM in your AI client
Folk doesn't ship first-party MCP yet, but NimbleBrain's community-maintained mcp-folk server at github.com/NimbleBrainInc/mcp-folk is the working integration today. It uses stdio transport with FOLK_API_KEY authentication, exposes Folk's core surface (contacts, companies, notes, reminders, interaction logging), and slots into Claude Desktop alongside any other MCP server you already run.
This page is the operator-grade walkthrough: the 5-10 minute setup, five concrete workflows that match Folk's relationship-led motion, the key-hygiene gotchas (long-lived API keys are not OAuth), and the realistic comparison to first-party hosted MCPs from Attio and HubSpot.
- Transport
- stdio Local Node process via npx
- Auth
- FOLK_API_KEY Long-lived key, store as env var
- Setup time
- 5-10 min One config edit + restart Claude
- Maintainer
- NimbleBrain Community-built, official-adjacent
TL;DR
Want to try Folk?
Folk + MCP is the relationship-led CRM motion in an AI client
Tag-and-list CRM built for solopreneurs, agencies, and small sales teams. Free tier available; AI Assistants bundle in the paid tiers; community MCP from NimbleBrain ships today.
Start with Folk →Affiliate link — StackSwap earns a commission if you sign up for Folk. We only partner with tools we'd recommend anyway.What Folk MCP exposes
Folk's CRM is intentionally narrow — tags, lists, contacts, companies, notes, reminders, interaction logs. There's no relational object schema (no custom objects like Attio), no native sequencing engine (Folk pairs with Smartlead/Instantly for cadence), no pipeline-stage workflow engine (Folk's pipeline view is lighter than HubSpot's). The MCP surface mirrors that narrow product: it exposes the operations that map to Folk's actual feature set, no more.
- Contacts. Read, search by tag/list/company, create, update, assign tags, add to lists.
- Companies. Read, search, create, link to contacts.
- Notes. Read interaction history on a contact, append new notes.
- Reminders. Read upcoming/overdue, create new reminders with due dates.
- Interactions. Log calls, meetings, emails against a contact with timestamp + context.
Five workflows that match Folk's motion
1. Weekly interaction-log summary
Prompt Claude with "pull all interactions from the past 7 days across my Investors list, group by company, and surface anyone I haven't touched in 30+ days." Claude routes the query through Folk MCP, aggregates the response, returns a structured Monday-morning digest. Replaces the manual "let me check each contact card" ritual that eats 30 minutes a week.
2. Bulk-tag cleanup from a LinkedIn scrape
Paste a list of new contacts captured via Folk's Chrome extension or a LinkedIn export into Claude. Ask Claude to assign matching Folk tags ("Founders 2026", "AI/ML", "EU") and add them to the right lists based on their LinkedIn descriptions. The LLM judgment replaces the manual tag-assignment step; the MCP write back to Folk is one operation per contact.
3. Follow-up reminder triage
"Show me all reminders due this week. For each, pull the contact's last 3 interactions and draft a short follow-up message grounded in that context." Claude fans out via the Folk reminders endpoint, pulls per-contact interaction history, drafts personalized outreach. You review, tweak, send. The drudgery part disappears.
4. Prep-for-meeting context
Drop a calendar event title or paste a meeting invite into Claude. Ask it to look up the contact in Folk, pull their interaction history + linked company + active reminders, and write a one-paragraph briefing. Replaces the "let me search Folk for this person" tab-switch right before every call.
5. Cross-list move based on engagement
"Move everyone in my Prospects list who's responded twice into the Active Discussion list, and tag them as 'engaged_q2'." Claude reads the Prospects list, filters by response count from interaction logs, executes the moves and tags. Twenty minutes of manual list-management becomes a single prompt.
Setup walkthrough
Folk MCP runs as a stdio Node process configured in your Claude Desktop config file. Path varies by OS: ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%/Claude/claude_desktop_config.json on Windows.
- Install Node.js if you don't have it (18+ recommended).
- Generate a Folk API key. Folk Settings → Developer → API Keys. Create a key scoped to a dedicated AI user (not your admin account). Copy the key.
- Edit your Claude Desktop config. Add an entry under
mcpServersthat invokes the mcp-folk package via npx with FOLK_API_KEY set as an env var. The repo README at github.com/NimbleBrainInc/mcp-folk has the exact snippet. - Restart Claude Desktop. The MCP tools from Folk appear in your next session.
- Verify connectivity. Ask Claude "list my recent Folk contacts" — if you see real data, you're wired.
The API-key hygiene gotcha
OAuth-based MCPs (Apollo, HubSpot, Attio) handle the secret-management burden for you. API-key MCPs like Folk put that burden on you. Three rules to follow:
- Dedicated key for AI work. Generate a new FOLK_API_KEY specifically for Claude integration. If you ever need to revoke, you don't break other integrations.
- Env var, not plaintext. In your config, reference the key via env block, not pasted directly in the command argument. Keeps the secret out of process listings.
- Don't commit the config. If you sync your Claude Desktop config across machines via Git, make sure the file with secrets is in
.gitignoreor that secrets reference shell env vars set outside the repo.
When Folk MCP doesn't unlock value
If you don't run daily orchestration through Claude / Cursor / ChatGPT and your team uses Folk only through the UI + AI Assistants feature, the MCP layer doesn't add anything to your motion. Folk's native AI Assistants do a lot of the same in-product work (drafting follow-ups, suggesting tags, summarizing interactions) without the 5-10 minute setup. Evaluate Folk on its core fit (tag-and-list CRM, LinkedIn capture, fast onboarding); add the MCP only if your AI client is part of your daily workflow.
Want to try Folk?
Folk's relationship-led CRM + community MCP is the right shape for solo/agency motion in 2026
Tag-and-list model, LinkedIn capture extension, Folk AI Assistants in-product, and NimbleBrain's mcp-folk for Claude/Cursor power users. Pair with Smartlead or Instantly MCP for cadence.
Start with Folk →Affiliate link — StackSwap earns a commission if you sign up for Folk. We only partner with tools we'd recommend anyway.FAQ
Related reading
- Folk MCP review — full operator analysis of the NimbleBrain community MCP
- Folk MCP vs Zapier — when each wins for CRM workflows
- Folk — full operator review of the relationship-led CRM
- Is Folk worth it in 2026? — operator-narrative buyer guide
- Best Folk alternatives 2026 — Attio, HubSpot, and the rest
- StackSwap MCP — the cross-vendor GTM meta-layer for stack-level decisions
- Best MCP servers for B2B SaaS operators 2026
- What is MCP for B2B SaaS operators
Canonical URL: https://stackswap.ai/folk-mcp-claude-integration. Disclosure: StackSwap is a Folk affiliate. The structural read above is the same operator analysis we'd give a GTM operator evaluating Folk cold against Attio and HubSpot.