Twenty MCP
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Twenty MCPfind all companies in the tech industry created this month"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Twenty MCP
A remote MCP (Model Context Protocol) server that connects Claude to a Twenty CRM workspace, deployed on Cloudflare Workers with OAuth for a one-click team install.
What it does
Exposes 8 generic, schema-driven tools that work on any Twenty object (Person, Company, Opportunity, or any custom object). The MCP introspects Twenty's metadata API at runtime — you never need to update the MCP when you add fields or objects.
Tools
list_objects,describe_object— discover what's in the CRMfind_records,get_record— query with filter/orderBy/paginationcreate_record,update_record,delete_record— mutations (only when connected in write mode)run_graphql— escape hatch for raw metadata/graphqlget_primer— org-specific domain context + live schema snapshot
Resources (auto-loaded by Claude on session start)
twenty://primer— organisation context merged with a compact schema snapshottwenty://api/info— connector status and current scopes
Install (team member, one-click)
In Claude → Settings → Connectors → Add custom connector
URL:
https://<your-worker>.workers.dev/mcpClaude redirects you to a consent page. Paste your personal Twenty API key (from Twenty → Settings → Developers → Generate API key).
Choose permission (read-only or read+write) and optional object scopes.
Done. Your key is stored encrypted in Cloudflare KV, bound to your session.
Changes you make in Twenty are attributed to your Twenty user, not to a shared service account.
Deploy (admin, first time)
Click the deploy button above. Cloudflare will fork the repo and prompt for secrets.
Set the required secrets:
COOKIE_ENCRYPTION_KEY—openssl rand -hex 32TWENTY_BASE_URL— your Twenty instance URL (e.g.https://crm.example.com)
(Optional) Set
ADMIN_TOKEN—openssl rand -hex 32— enables the/admin/*endpoints for primer management.Note the deployed URL, e.g.
https://twenty-mcp.your-subdomain.workers.dev.Share
<url>/mcpwith the team.
For Claude managed team: register <url>/mcp once in the Claude team admin console — it will appear in every team member's connector list. Each member still completes the one-time consent page to paste their own Twenty API key.
Customising the primer
The twenty://primer resource gives Claude context about your CRM before any tool calls. It contains two parts:
Organisation context — a markdown doc describing your domain model, custom objects, business rules, and conventions. Things introspection can't capture (e.g. "Object A and Object B are independent — don't infer one from the other").
Schema snapshot — auto-generated from Twenty's metadata API, cached for 1 hour.
Out of the box, part (1) is a generic Twenty template. To upload your org-specific context:
# Set ADMIN_TOKEN first if you haven't:
# wrangler secret put ADMIN_TOKEN
# Upload your context markdown:
curl -X PUT https://<your-worker>.workers.dev/admin/primer \
-H "Authorization: Bearer <your-admin-token>" \
-H "Content-Type: text/markdown" \
--data-binary @path/to/your-context.md
# Verify it's loaded:
curl https://<your-worker>.workers.dev/admin/primer \
-H "Authorization: Bearer <your-admin-token>"
# Revert to the bundled default:
curl -X DELETE https://<your-worker>.workers.dev/admin/primer \
-H "Authorization: Bearer <your-admin-token>"The context markdown should describe: what your organisation does, what each custom object means and how they relate, tiering/classification models, naming conventions, and any "do this / don't do that" rules for AI. See src/primer/default-context.md for the template structure.
Local development
npm install
cp .dev.vars.example .dev.vars
# edit .dev.vars — set COOKIE_ENCRYPTION_KEY, TWENTY_BASE_URL, and optionally ADMIN_TOKEN
npm run dev # wrangler dev on http://localhost:8787
npm run typecheckTo connect a local Claude Desktop to the dev worker, add http://localhost:8787/mcp as a connector.
How auth works
Twenty has no upstream OAuth provider — authentication is via per-workspace API keys. So:
The Worker runs its own OAuth 2.1 endpoint (required by Claude connectors).
During the OAuth consent step, the user pastes their Twenty API key into an HTML form.
The Worker validates the key against Twenty's
/metadataendpoint, then stores{twentyApiKey, mode, allowedObjects, label}as encrypted OAuth props.Every subsequent MCP tool call has the user's key available via
this.props.
This means the MCP is OAuth on the outside (for Claude) and API key on the inside (for Twenty).
Scoping
Each connection can be narrowed at install time:
Mode: read-only hides
create_record/update_record/delete_record.Allowed objects: comma-separated list to restrict to specific objects.
Object-level permissions are also enforced by Twenty itself via the role attached to the user's API key — belt and braces.
Architecture
Claude ↔ OAuth 2.1 ↔ Worker ↔ REST+GraphQL ↔ Twenty workspace
│
├─ McpAgent Durable Object (per session)
├─ OAUTH_KV (token store, schema cache, primer)
└─ twenty://primer (org context + live schema)License
Apache-2.0. Based on the same architecture as fathom-mcp.
This server cannot be installed
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/High-Impact-Athletes/hia-twenty-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server