Crustdata for Dify

People, company and job data in Dify — 800M+ profiles and 200M+ companies, with employment
history, titles and seniority, education, skills, location, headcount over time, funding,
technologies, job postings, social posts, work emails and phone numbers, plus live web search
and page fetch.

The plugin is a thin wrapper over the Crustdata MCP server
at . It exposes that server's three meta-tools as three Dify
tools, and adds nothing of its own — the tool inventory, the schemas and the steering all come
from the API at call time, so the plugin does not drift as the API changes.

Setup

1. Install the plugin.
2. Open its settings and paste your Crustdata API key (find it in the
   Crustdata dashboard).

The key is sent as a bearer token to the MCP endpoint on each call. Nothing is stored by the plugin.

Tools

How it is meant to be used

Crustdata's MCP surface is Code Mode: instead of one tool per endpoint, an agent writes a small
script that calls the endpoints it needs and returns a narrow result. So an agent should call
once, for the tools it plans to use, then do the whole task in a single
.

A script must open with a source-labelled query comment, is plain JavaScript (never TypeScript),
and reaches the API through one primitive — . A query builder (, , ,
, …) and result shapers (, , , , …) are preloaded, no import:

never throws — it returns or
, so a failed call does not abort the script. Note the
whitelist: a result carries only the groups you list, so a group you forgot to ask for
reads back .

Set Dry run to rehearse a script against committed fixtures for zero credits before paying for
a long one.

Agent vs. workflow

In an Agent node, add all three tools and let the model discover the surface: it will call
, then , then .

In a Workflow, alone is usually enough — write the script once in the node and pass
run-specific values through Inputs rather than interpolating them into the code.

Write Inputs as JSON text, e.g. , and reference it as in
the script. Dify parses that text into the object the script receives; a value supplied as a
structured constant rather than JSON text arrives empty, and the script then sees as
with no error to explain why.

Timeouts

A script that fans out across many API calls runs longer than an ordinary request. The plugin
allows up to 600 seconds; cap what comes back with Max output bytes on the tool if
results are large.

Development

Live debugging

Dify can run the plugin straight from your machine, so you can edit and re-test without
packaging or reinstalling anything.

The first run creates and stops. Open Dify → Plugins, click the bug icon next to
Install Plugin, and paste the key from that dialog into . Run the script
again and the plugin appears in your workspace, tagged as debugging.

• defaults to Dify Cloud (). For self-hosted Dify set
  , and make sure the daemon actually exposes that port.
• The plugin lives only as long as the process does — Ctrl-C and it vanishes from the workspace.
• There is no hot reload. After editing, restart the script.
• Every invocation Dify sends is logged to the terminal, which is the point: you see the real
  tool parameters a model produced, not what you assumed it would.

To verify the plugin against the live Crustdata API without involving Dify at all:

That checks the handshake, all three tools, a dry run and a live run. It spends no credits beyond
the one free lookup.

Source and support

Source: github.com/crustdata/dify-plugin. The plugin is
developed in Crustdata's internal repository, so send changes to
[email protected] rather than opening a pull request.

Questions, bugs and API key problems: [email protected].