data-aggregator-mcp
by musharna
README.md
# data-aggregator-mcp
data-aggregator-mcp is an MCP server that lets AI agents search 21 public
research-data sources at once and get back one deduplicated list.
[](https://github.com/musharna/data-aggregator-mcp/actions/workflows/ci.yml)
[](https://pypi.org/project/data-aggregator-mcp/)
[](https://pypi.org/project/data-aggregator-mcp/)
[](https://github.com/musharna/data-aggregator-mcp/blob/main/LICENSE)
[](https://doi.org/10.5281/zenodo.21636332)
[](https://glama.ai/mcp/servers/musharna/data-aggregator-mcp)
<!--
mcp-name: io.github.musharna/data-aggregator-mcp
-->
<p align="center">
<img src="https://raw.githubusercontent.com/musharna/data-aggregator-mcp/main/examples/assets/demo.svg"
alt="Live calls over stdio: a search for transcriptome data on Orobanche aegyptiaca that NCBI Taxonomy expands to Phelipanche aegyptiaca and that returns hits from SRA, BioProject, PubMed, OpenAIRE and DataCite; resolving one SRA hit to two FASTQ files with sizes and md5 checksums; fetching an OpenML ARFF file with its md5 verified; and SQL over a remote CSV on Hugging Face without downloading it"
width="820">
</p>
## Install
Claude Code:
```bash
claude mcp add data-aggregator -- uvx data-aggregator-mcp
```
One-click install for VS Code or Cursor:
[](https://insiders.vscode.dev/redirect/mcp/install?name=data-aggregator&config=%7B%22command%22%3A%22uvx%22%2C%22args%22%3A%5B%22data-aggregator-mcp%22%5D%7D)
[](https://cursor.com/install-mcp?name=data-aggregator&config=eyJjb21tYW5kIjoidXZ4IiwiYXJncyI6WyJkYXRhLWFnZ3JlZ2F0b3ItbWNwIl19)
Any other MCP client: run `uvx data-aggregator-mcp` as a stdio server.
<details>
<summary>Claude Desktop, pip, and the <code>operate</code> extra</summary>
Claude Desktop (`claude_desktop_config.json`) and most other clients:
```json
{
"mcpServers": {
"data-aggregator": {
"command": "uvx",
"args": ["data-aggregator-mcp"],
"env": { "NCBI_API_KEY": "your-optional-key" }
}
}
}
```
With pip:
```bash
pip install data-aggregator-mcp
data-aggregator-mcp # or: python -m data_aggregator_mcp
```
`operate` (SQL and previews over remote files) needs an optional extra:
`pip install "data-aggregator-mcp[operate]"`.
</details>
## What you can ask
**"Find transcriptomes for Orobanche aegyptiaca."** The species was renamed; NCBI
Taxonomy maps it to *Phelipanche aegyptiaca*, and the sources that accept synonyms
match both names (the demo above).
**"What data came out of this paper?"**
```text
resolve("pubmed:40098680")
links geo:GSE284240, bioproject:PRJNA1198054, and 18 sra: experiments
files PMC11910882.xml (open-access full text)
```
**"Can I train a model on this dataset?"**
```text
resolve("hf:scikit-learn/iris", use="ml-training")
license_compat ALLOW: "CC0-1.0 grants the permission(s) required for
ml-training: commercial-use, modifications"
```
The verdict is read from the record's license metadata; it is advice, not a legal
opinion. `resolve` also renders citations in any CSL style, checks Crossref for
retractions, scores FAIRness, and exports Croissant or RO-Crate. `relate` reports
how a set of records connect (a shared accession, DOI or explicit link).
## How a search runs
<p align="center">
<img src="https://raw.githubusercontent.com/musharna/data-aggregator-mcp/main/docs/assets/architecture.svg"
alt="How one search runs: NCBI Taxonomy maps Orobanche aegyptiaca to its current name, Phelipanche aegyptiaca, and the sources that accept synonyms search both names; the query goes to 21 sources in parallel (11 archives, 9 omics, 1 literature); records that share a DOI become one, keeping the copy that can be downloaded; a record can then be resolved to files and checksums, fetched with the checksum verified where the source publishes one, or queried with SQL without downloading it"
width="820">
</p>
A source that fails is named in the result's `errors`, never dropped silently.
What each source can fetch, and which checksum it is verified against:
[docs/sources.md](https://github.com/musharna/data-aggregator-mcp/blob/main/docs/sources.md).
## What the benchmark shows
An agent using this server was compared with the same agent using web search, on held-out
dataset-finding tasks scored by recall against independently built answer keys, 3 runs per
task. Mean recall, server against web search: Claude Haiku 5.5 0.47 against 0.32 (higher on
8 of 10 tasks), Sonnet 5.5 0.54 against 0.52 (5 of 10), Opus 5.5 0.72 against 0.58 (6 of 10).
With Haiku the server met round 5's pre-registered "better" rule (higher on at least 8 of 10
tasks, interval above 0) but not the stricter bar for a win across model tiers (9 of 10,
Holm-adjusted sign test, p = 0.22); Sonnet and Opus met neither. With Opus it returned fewer wrong
entries and answered in about a third of the time; with Sonnet it was faster on 9 of 10
tasks; with Haiku it cost less on 9 of 10.
The two latest Opus 5.5 rounds (2026-10-19, rules written before the runs):
| Tools given to the agent | Round 4 (5 tasks) | Round 5 (10 tasks) |
| --- | :---: | :---: |
| this server | 0.66 | 0.72 |
| web search | 0.66 | 0.58 |
| ToolUniverse | 0.59 | 0.63 |
| no tools (memory only) | 0.02 | 0.09 |
In round 5 this server was higher on 6 of 10 tasks, where the pre-set rule needed 8.
Web search hit the 15-minute limit in 11 of 30 runs, and a run that times out scores
0. In runs that finished, web search found more (0.90). The server answered in 203 s
against 599 s (median), with fewer tool calls and fewer wrong entries (precision 0.95
against 0.62). It cost more per run than web search ($2.08 against $1.51,
API-equivalent).
We chose the tasks and ran the benchmark ourselves; an outside re-run is welcome. Method
and controls:
[benchmarks/h2h](https://github.com/musharna/data-aggregator-mcp/blob/main/benchmarks/h2h/README.md);
the Haiku and Sonnet runs:
[2026-10-20 re-run](https://github.com/musharna/data-aggregator-mcp/blob/main/benchmarks/h2h/README.md#haiku-and-sonnet-re-run-2026-10-20).
## Compared with other servers
| Server | Searches | One search | Checksum | Remote SQL | License check |
| --- | --- | :---: | :---: | :---: | :---: |
| **data-aggregator-mcp** | 21 sources: data archives, omics, papers | yes, DOI dedup | where published | yes | yes |
| [Mobus](https://github.com/mobus-ai/Mobus) | 20 general and ML data platforms | yes | — | row preview | yes |
| [paper-search-mcp](https://github.com/openags/paper-search-mcp) | papers: arXiv, PubMed, OpenAlex, Crossref and more | yes, deduplicated | — | — | — |
| [ToolUniverse](https://github.com/mims-harvard/ToolUniverse) | 1000+ tools: models, datasets, APIs, packages | papers | — | — | — |
| [BioMCP](https://github.com/genomoncology/biomcp) | genes, variants, trials, drugs, proteins, papers | yes (`search all`) | — | — | — |
— means the project's README doesn't describe it (READMEs read 2026-10-04).
Where they are ahead: BioMCP reaches clinical trials, ChEMBL and AlphaFold;
ToolUniverse covers far more ground overall; paper-search-mcp covers far more
literature sources; Mobus also compares datasets and checks schema compatibility.
More: [docs/POSITIONING.md](https://github.com/musharna/data-aggregator-mcp/blob/main/docs/POSITIONING.md).
## Reference and development
Every tool and parameter, the HTTP transport (`--transport http`) and the
environment variables:
[docs/reference.md](https://github.com/musharna/data-aggregator-mcp/blob/main/docs/reference.md).
Sources that need every word of a query, and how they are searched:
[every-word sources](https://github.com/musharna/data-aggregator-mcp/blob/main/docs/reference.md#every-word-sources).
Tests, linting and the live API probes:
[CONTRIBUTING.md](https://github.com/musharna/data-aggregator-mcp/blob/main/CONTRIBUTING.md).
## License
MIT. See [LICENSE](https://github.com/musharna/data-aggregator-mcp/blob/main/LICENSE).
TDQS
A4.4/5.0
Scored across 6 tools
Disambiguation5/5
Each tool has a distinct role: search discovers resources, resolve expands a known id, fetch downloads files, list_sources introspects coverage, operate analyzes remote tabular files without downloading, and relate compares resource ids. The adjacent file-related operations (resolve/fetch/operate) are explicitly scoped by description, so an agent can tell them apart.
Naming Consistency4/5
Five tools are bare lowercase verbs (search, resolve, fetch, operate, relate) and one is a snake_case verb_noun (list_sources), so the pattern is not perfectly uniform. It remains predictable and readable, with no chaotic mixing of camelCase or vague synonym pileups.
Tool Count5/5
Six tools is well-scoped for a read-only multi-archive research-data aggregator, and each tool covers a distinct workflow stage rather than duplicating another. The surface is neither thin nor bloated, and no obvious filler tool is present.
Completeness5/5
The set covers discovery, detailed record resolution, file download, source capability introspection, remote tabular inspection, and cross-resource relation hints, which is strong lifecycle coverage for this domain. Write/update operations are not expected for a read-only aggregation server, and remaining gaps look like minor conveniences rather than dead ends.
Maintenance
ActivityActive
ResponsivenessMixed handling