
Crystallize MCP
io.github.HayoDevPublished by GitHub account HayoDevv0.2.0Updated Oct 25, 2026
Crystallize headless commerce tools for AI agents via Model Context Protocol
Overview
Local MCP server that gives an AI assistant read and write access to a Crystallize headless commerce tenant: catalogue, shapes, orders, customers and content.
- What you can do
- Browse, fetch and keyword-search catalogue items, and list product variants with pricing and stock; every response includes a deep link back to the Crystallize UI. Explore shapes and their queryable fields, and browse items of a shape with filters, pagination and field selection. Read shapes and tenant configuration, including available languages (requires auth). List and inspect orders and customers with pagination (requires auth). Create items and update single component values, including nested content chunks, in draft only.
- When to use it
- Useful when an assistant should work inside a Crystallize store: looking up products and content, answering questions about orders and customers, or drafting new items and component edits that a person then reviews and publishes in the Crystallize UI.
- Requirements
- Runs locally via npx with Node.js. CRYSTALLIZE_TENANT_IDENTIFIER is required; catalogue and discovery tools work without auth. PIM tools (shapes, tenant info, orders, customers) need CRYSTALLIZE_ACCESS_TOKEN_ID and CRYSTALLIZE_ACCESS_TOKEN_SECRET, or CRYSTALLIZE_STATIC_AUTH_TOKEN. Write tools need CRYSTALLIZE_ACCESS_MODE set to write or admin and a token with write permissions. Network access to Crystallize is required.
Installation
In SourceWeft
- Open Crystallize MCP in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
Crystallize MCP
[npm version] [npm downloads] [license] [node]
MCP server for Crystallize headless commerce. Gives AI agents read and write access to your catalogue, products, shapes, orders, customers, and tenant config — with deep links back to the Crystallize UI, dry-run safety for mutations, and PII masking for customer data.
Works with Claude Code, Claude Desktop, Cursor, Windsurf, Copilot, and any MCP-compatible client.
Getting started
Setup wizard
The interactive wizard handles config, auth tokens, keychain storage, and PII mode in one step:
Manual config
Standard MCP config (works in any client):
Add CRYSTALLIZE_ACCESS_TOKEN_ID and CRYSTALLIZE_ACCESS_TOKEN_SECRET to the env block for PIM tools (shapes, orders, customers). See Authentication.
Claude Code (CLI)
Use --scope project to write to .mcp.json (shared with your team) or --scope user for personal use across all projects.
Claude Desktop
Add the standard config to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows).
Cursor
Add the standard config to your Cursor MCP settings (~/.cursor/mcp.json or the project-level .cursor/mcp.json).
VS Code / GitHub Copilot
Add the standard config to .vscode/mcp.json in your project root.
Windsurf
Add the standard config to ~/.codeium/windsurf/mcp_config.json.
Gemini CLI
Add the standard config to ~/.gemini/settings.json.
JetBrains AI Assistant
Add the standard config to .junie/mcp.json in your project root.
Warp
Add the standard config to ~/.warp/mcp.json.
Raycast
Open "Install MCP Server" in Raycast and fill in:
- Command:
npx - Arguments:
-y @hayodev/crystallize-mcp@latest - Environment: add
CRYSTALLIZE_TENANT_IDENTIFIERand your token vars
Or copy the standard config JSON above before opening the command — Raycast will auto-fill the form.
From source (maintainers)
The --local flag is for developing crystallize-mcp itself. It writes .mcp.json pointing to the local build output — run this from the repo root only:
Or point your MCP client directly at the built entry point:
Tools (16)
Catalogue (4 tools)
Discovery (3 tools)
Shapes & Tenant (3 tools, requires auth)
Orders (2 tools, requires auth)
Customers (2 tools, requires auth)
Content (2 tools, requires auth + write mode)
Write tools
Write tools require CRYSTALLIZE_ACCESS_MODE=write (or admin) and a token with write permissions.
Dry-run mode
Set CRYSTALLIZE_DRY_RUN=true to preview mutations without executing them. The response shows exactly what would change — the mutation payload, before/after values, and a deep link to the item:
Example prompts
Create an item:
"Create a new blog post under /blog using the article shape with title 'Getting Started'"
Update a top-level component:
"Find the item at /products/summer-collection and update its description to 'New summer arrivals'"
Update a component inside a content chunk:
"Get the item at /articles/my-post, then update hero.title to 'Updated Headline' and give me the deep link to review the draft"
Update with change summary:
"Get the item at /articles/guides/my-guide, update its title component to 'New Guide Title', give me the deep link, and show a table of which fields in the chunk changed vs remained unchanged with before/after values"
The agent will update the target component in draft only, preserve all sibling components in the chunk, and return a summary like:
No publishing happens — you review the change in the Crystallize UI via the deep link and publish when ready.
Authentication
Catalogue and Discovery tools work without auth — just set CRYSTALLIZE_TENANT_IDENTIFIER.
For PIM tools (shapes, tenant info, orders, customers), create an access token at:
https://app.crystallize.com/{tenant}/en/settings/access-tokens
Note: Crystallize tokens inherit permissions from the user who created them. To restrict an agent to read-only access, generate the token under a user with a read-only role. See Crystallize Roles.
Environment variables
PII mode (opt-in)
By default all customer and order data is returned as-is (full). Set CRYSTALLIZE_PII_MODE to opt in to data minimisation:
Applies to list_customers, get_customer, list_orders, and get_order. No effect on catalogue or shape tools.
Relevant for teams handling real customer data, GDPR Article 25 compliance (data minimisation by design), or environments where the AI doesn't need raw contact details to do its job. A common pattern is masked on production and full on dev.
Audit log (opt-in)
Set CRYSTALLIZE_AUDIT_LOG to an absolute file path to enable structured logging of every tool call:
One JSON line per call — timestamp, tool name, params, result (ok/error), and tenant. Write tools also log mutation metadata (before/after state) for audit trails.
Note: Params are logged as-is and may contain PII — for example, a
searchTermof[email protected]or acustomerIdentifier. Treat the audit log file as sensitive data and restrict access accordingly. Param scrubbing is on the roadmap but not yet implemented.
Easy to pipe into log aggregators (Datadog, CloudWatch, Splunk) — but ensure your pipeline handles the file with appropriate access controls.
Keychain storage (optional)
The setup wizard (npx @hayodev/crystallize-mcp --setup) can store tokens in the OS keychain (macOS Keychain, Windows Credential Manager, or libsecret on Linux) so they never appear as plain text in config files. The MCP server resolves credentials from the keychain automatically at startup — no extra configuration needed.
This is useful when:
- Your
.mcp.jsonis committed to git — keychain keeps tokens out of the repository - You prefer not to have secrets in plain text config files — the config only needs
CRYSTALLIZE_TENANT_IDENTIFIER
When using --setup --global with Claude Code, this applies only if keychain storage is available and you opt into it: in that case, the wizard runs claude mcp add with only the non-secret env vars, and tokens are read from the keychain at runtime so they do not appear in Claude Code config. If keychain storage is unavailable or you choose not to use it, the wizard passes token env vars to claude mcp add, and Claude Code may store them in plain text.
Access mode
CRYSTALLIZE_ACCESS_MODE controls which tools the MCP server registers at startup:
read(default) — read-only tools onlywrite— includes content creation and component updates (with dry-run support)admin— full access including shape modifications and tenant config
Deep links
Every response includes clickable links to the Crystallize UI:
- Items →
app.crystallize.com/@{tenant}/{language}/catalogue/{type}/{itemId} - Shapes →
app.crystallize.com/@{tenant}/{language}/settings/shapes/{identifier} - Orders →
app.crystallize.com/@{tenant}/{language}/orders/{orderId}
The language segment is automatically set from the tenant's default language, bootstrapped at server startup — no configuration needed. Tools that accept a language parameter (catalogue, search) use the requested language in both the API call and the generated link.
Development
License
MIT
Source: README.md at commit 68282f3
Tools
0Version history
1- v0.2.0LatestSep 16, 2026
Score
50% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts