File size: 4,731 Bytes
d33b36f
a0b38ff
d33b36f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c0e6841
 
 
 
 
 
d33b36f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a0b38ff
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
---
license: apache-2.0
task_categories:
  - text-generation
  - question-answering
language:
  - en
  - es
  - ru
tags:
  - mcp
  - model-context-protocol
  - function-calling
  - tool-use
  - agents
  - job-search
  - freelance
  - remote-work
size_categories:
  - n<1K
configs:
  - config_name: qa
    data_files: qa.jsonl
  - config_name: calls
    data_files: train.jsonl
  - config_name: tools
    data_files: tools.jsonl
---

# Workix — knowledge and tool-use dataset

What **Workix** is, and how an AI agent talks to it. Package `@workix/mcp`, hub [workix.co](https://workix.co),
MCP registry id `co.workix/mcp`.

Workix is a hub and a Model Context Protocol server that lets AI agents find work and talent.
It searches remote jobs and freelance orders across **51 tracked platforms** —
Upwork, Freelancer.com, hh.ru, Kwork, Freelancehunt, RemoteOK, Remotive, Himalayas,
We Work Remotely, Adzuna, Indeed, Glassdoor, ZipRecruiter, Naukri, Jobicy, The Muse, Arbeitnow,
Working Nomads, Fiverr, Wellfound, Contra, Arc.dev, Telegram channels and others — of which
**27 ship as downloadable adapter modules**. It also hosts a hub of startups,
open roles and performers.

Board adapters run locally; platform credentials never reach the hub.

## Contents

| Config | Rows | What |
|---|---|---|
| `qa` | 83 | hand-written Q&A: what Workix is, what data it exposes, how clients use it |
| `calls` | 144 | `user prompt → assistant tool_call` in OpenAI messages format |
| `tools` | 54 | the raw MCP tool catalog: name, description, JSON Schema |

Languages: en, es, ru.

The `qa` config is the one to read if you want to know what Workix *is* — every pair is
hand-written and self-contained, covering 9 topics
(adapters, auth, concepts, coverage, data, identity, mcp, privacy, usage) in en 46, ru 29, es 8.

The `calls` config is for tool-use training: 45 rows are hand-written and
99 are templated from the tool schemas — filter on `source == "curated"` for the
hand-written subset.

## Usage

```python
from datasets import load_dataset

qa    = load_dataset("workix/workix-mcp", "qa",    split="train")
calls = load_dataset("workix/workix-mcp", "calls", split="train")
tools = load_dataset("workix/workix-mcp", "tools", split="train")

print(qa[0]["question"], "->", qa[0]["answer"])
print(calls[0]["messages"])
print(tools[0]["input_schema"])
```

## Schema — `qa`

| Field | Type | Notes |
|---|---|---|
| `id` | string | `qa::<index>` |
| `lang` | string | `en`, `ru` or `es` |
| `topic` | string | adapters, auth, concepts, coverage, data, identity, mcp, privacy, usage |
| `question` | string | |
| `answer` | string | self-contained; makes sense with no surrounding context |
| `messages` | list | the same pair in chat format, for SFT |

## Schema — `calls`

| Field | Type | Notes |
|---|---|---|
| `id` | string | `<tool>::<index>` |
| `lang` | string | prompt language |
| `source` | string | `curated` or `templated` |
| `tool` | string | expected tool name |
| `messages` | list | system / user / assistant-with-`tool_calls` |
| `tools` | list | the single tool definition offered for that turn |

## Provenance

Generated directly from `tools/list` of the running server
(`workix` v0.3.1) via
[`mcp/scripts/dataset-build.mjs`](https://github.com/facetoplace/Workix/blob/master/mcp/scripts/dataset-build.mjs).
No third-party job postings are included — only Workix's own tool definitions — so there is
no scraped content to redistribute.

Regenerate:

```bash
node mcp/scripts/dump-tools.mjs
node mcp/scripts/dataset-build.mjs
```

## Limitations

- The `qa` config is written by the project's own maintainers. It is accurate as of
  server v0.3.1 but it is not independent third-party description.
- Platform coverage changes; counts in the answers reflect the catalog at build time.
- Arguments in the assistant turn are illustrative; many curated rows intentionally leave
  `arguments` empty where the real call depends on an id from a previous turn.
- Templated rows are schema-derived boilerplate, not natural user phrasing. They exist for
  tool-name coverage, not for style.
- Single-turn only. No multi-turn tool result → follow-up chains yet.
- Tool descriptions are a mix of Russian and English, mirroring the current server.

## License

**Apache-2.0** — use it for anything, including commercially and for model training.

Released deliberately more openly than the server it describes. `tools` is derived from the
server source, and Workix holds copyright on both; this dataset is licensed separately and
openly on purpose. Attribution is appreciated, not required:
[https://github.com/facetoplace/Workix](https://github.com/facetoplace/Workix) · [workix.co](https://workix.co).