zenodo-mcp-server

Public Hosted Server: https://zenodo.caseyjhand.com/mcp

Overview

Datasets, software releases, and publications from Zenodo, CERN's open research repository. Search deposits with filters for funder, grant, community, license, and file type; resolve any Zenodo DOI, concept DOI, or URL to its record; walk a deposit's versions; and list, open, and preview its files, including members of .zip archives.

Tools

The published zenodo-guide covers selecting a cited release, resolving funding identities, and checking inspection coverage. Plugin installs include it; MCP clients can discover it through the Skills extension and its skill resources.

Related MCP server: mcp-figshare

Capability reference

zenodo_search_records tool

• Keyword query (terms OR-ed unless joined with AND, quoted phrases, field syntax such as metadata.title:"…") plus resource_type, community, funder, award, creator_orcid, file_type, license, access_status, and published_from / published_to

• Up to 25 hits per page; only the first 10,000 matches are reachable (result_window_exceeded past that); latest versions only unless all_versions is true

• sort: bestmatch, newest, oldest, mostviewed, mostdownloaded, updated-desc, updated-asc (default bestmatch with a query, newest without)

• Each hit carries record and concept ids and DOIs, type, version, creators, license ids, access, file totals, and unique views and downloads; facets count resource types, access statuses, file types, subjects, and years over the full match set

• A bare DOI or record URL as query fails as query_is_identifier; a community or funder Zenodo doesn't know fails as unknown_community / unknown_funder

• Invalid search syntax fails as query_syntax, naming the offending token:

  • With a filter other than community or all_versions set, any invalid syntax fails (climate AND AND model, metadata.title: with no value, 'single quotes', an unclosed [a TO b]), because Zenodo would search it as plain text and drop the filters. Without such a filter it runs as plain text and the notice says so

  • Constructs Zenodo cannot run fail with or without a filter, unless the query already runs as plain text: an unclosed / (escape a literal one as \/), a trailing ! (climate\!), a range bound ending in a backslash (double-quote a bound holding a space, ], or }: [a TO "b]"]), and a boost, fuzziness, or proximity value Zenodo cannot convert (climate^1.2.3, "x y"~0.5; write ^2, ~0.5, "x y"~2)

  • An unclosed " inside a word (12" vinyl; escape it as 12\") fails without a filter; with one, the phrase is closed at the end of the query and the closed query is checked instead

zenodo_get_record tool

• id takes a record id, a Zenodo DOI, a concept DOI or concept record id (resolves to the latest version), another DOI registered to a Zenodo record, or a zenodo.org / doi.org URL; input_kind and resolved_from report how it resolved

• Returns the description as plain text (up to 4,000 characters), creators and contributors with ORCIDs and ROR affiliations, rights, access and embargo, funding, related identifiers, communities, version position with latest_recid, usage counts, and the first 25 files

• Every capped field reports its total, and each cut except the file list (paged with zenodo_list_files) is listed in continuations with a cursor and the record id it is bound to; pass both back to get that field's next window (8,000 characters of text, or the next page of entries)

• citation_style: bibtex, csl-json, apa, chicago-author-date, harvard-cite-them-right, ieee, modern-language-association, or nature

• A miss returns found: false with miss_kind (not_found, deleted, restricted, not_on_zenodo) and guidance; a deleted record adds its removal tombstone

zenodo_list_versions tool

• Takes the same id forms as zenodo_get_record; a concept DOI and any version's DOI list the same series

• Newest first, up to 25 per page; reports total_versions, latest_recid, and the concept record id and DOI

• Each version carries its record id, DOI, version label, publication date, index, is_latest, file totals, and unique views and downloads

• Misses return found: false with the same miss_kind, guidance, and tombstone as zenodo_get_record

zenodo_list_files tool

• Manifest entries carry key, size, MIME type, MD5, download URL, previewable, and listable (a .zip); up to 200 per page via offset / limit

• archive_key lists the members of one .zip; Zenodo lists at most 1,000 files and directories per archive, flagged by upstream_truncated

• key_contains filters keys or member paths case-insensitively before paging

• Restricted and embargoed files return the access status and embargo date with no entries; unknown and deleted records fail as record_not_found / record_deleted

zenodo_read_file tool

• key from zenodo_list_files, plus archive_member to read inside a .zip without downloading it

• max_bytes from 256 (default 16,384), at most 65,536 per call (larger values return 65,536); continue a top-level file from next_offset; .zip members read from byte 0 only

• status is text, not_text, restricted, or empty; excerpts end on a line or character boundary

• Extensionless files typed application/octet-stream (LICENSE, Makefile) are returned only when their content is UTF-8 text

• Every result carries the deposit's rights and the download_url

zenodo_lookup_vocabulary tool

• vocabulary: communities, funders, awards, licenses, or resource_types; query by name, acronym, or keyword, or omit it to browse

• Each entry's filter_param and filter_value name the zenodo_search_records filter and the id to pass it

• funder scopes awards to one funder, as a ROR id, ROR URL, or Crossref Funder DOI

• Up to 25 entries per page; only the first 10,000 entries of a list are reachable, reported as reachable beside total

Features

Built on @cyanheads/mcp-ts-core: stdio and Streamable HTTP transports, pluggable auth (none / jwt / oauth), swappable storage (in-memory, filesystem, Supabase, Cloudflare KV/R2/D1), structured logging with optional OpenTelemetry tracing.

Zenodo-specific:

• Accepts a record id, Zenodo DOI, concept DOI, external DOI, or zenodo.org / doi.org URL in every tool that takes id; URLs are parsed locally and never fetched

• Search filters compose into Zenodo's query, so facet counts agree with total; community and funder values are checked before the search runs

• Separate request pacers for Zenodo's search and general rate-limit buckets, backed by its X-RateLimit-* headers, plus a process-local cache (records 5 min, searches 60 s, vocabularies 1 h)

• Byte-range file reads and .zip member reads, with binary detection before any text is returned

Agent-friendly output:

• Misses as data: zenodo_get_record and zenodo_list_versions return found: false with a typed miss_kind, next-step guidance, and a deleted record's tombstone

• Depositor-supplied text is labeled untrusted: descriptions render as quoted blocks, file content in a code fence, and HTML descriptions are converted to plain text

• Search responses echo the effectiveQuery sent to Zenodo and the appliedSort, and every paged tool reports totals with a next page or offset

• Fields Zenodo omits stay absent rather than defaulting to 0, '', or false

Getting started

Connect to the hosted instance, no install:

claude mcp add --transport http zenodo-mcp-server https://zenodo.caseyjhand.com/mcp

Or run it locally:

claude mcp add zenodo-mcp-server -- npx -y @cyanheads/zenodo-mcp-server

For other clients, add this to the MCP config file:

{
  "mcpServers": {
    "zenodo-mcp-server": {
      "command": "npx",
      "args": ["-y", "@cyanheads/zenodo-mcp-server"]
    }
  }
}

The hosted instance shares Zenodo's anonymous rate limit across callers. An optional Zenodo personal access token in ZENODO_ACCESS_TOKEN raises the rate limit of your own instance (-e ZENODO_ACCESS_TOKEN=your-token after the Claude Code server name, or an env entry). For Docker or a source build, see Running the server.

Configuration

See .env.example for the full list of optional overrides.

Rate limits

Zenodo rate-limits anonymous clients per IP address: 60 requests per minute and 2,000 per hour overall, and 30 per minute on record search. The server paces its own requests below those limits: 25 searches per minute, and 55 other requests per minute and 1,900 per hour. Every caller of one server process shares that budget. When it runs out, tools fail with rate_limited and a retryAfter in seconds.

With ZENODO_ACCESS_TOKEN set, Zenodo allows 100 requests per minute and 5,000 per hour, and the server paces other requests at 90 per minute and 4,800 per hour. Search stays at 25 per minute.

The token authenticates as the account that created it, so reads can see what that account can see, including restricted records it owns. On a shared or hosted server, create the token on a dedicated account that owns no restricted records.

Running the server

Local development

Requires Bun v1.4.0 or higher (or Node.js v24+).

• Clone and install:

  git clone https://github.com/cyanheads/zenodo-mcp-server.git
  cd zenodo-mcp-server && bun install
  cp .env.example .env   # optionally set ZENODO_ACCESS_TOKEN

• Build and run the production version:

  # One-time build
  bun run rebuild
  
  # Run the built server
  bun run start:http
  # or
  bun run start:stdio

• Run checks and tests:

  bun run devcheck  # Lints, formats, type-checks, and more
  bun run test      # Runs the test suite

Docker

docker run --rm -p 3010:3010 ghcr.io/cyanheads/zenodo-mcp-server:latest

The image defaults to HTTP transport, stateless session mode, and logs to /var/log/zenodo-mcp-server. To run it over stdio from an MCP client, use "command": "docker" with "args": ["run", "-i", "--rm", "-e", "MCP_TRANSPORT_TYPE=stdio", "ghcr.io/cyanheads/zenodo-mcp-server:latest"]. To build locally, docker build -t zenodo-mcp-server .; OpenTelemetry peer dependencies are installed by default — build with --build-arg OTEL_ENABLED=false to omit them.

Project structure

Development guide

See CLAUDE.md for development guidelines and architectural rules. The short version:

• Handlers throw, framework catches — no try/catch in tool logic

• Use ctx.log for logging; Zenodo responses are cached process-wide in the service, not in ctx.state

• Register new tools in src/mcp-server/tools/definitions/index.ts

• Wrap external API calls: validate raw → normalize to domain type → return output schema; never fabricate missing fields

Contributing

Issues are welcome. Run checks and tests before submitting:

bun run devcheck
bun run test

License

This project is licensed under the Apache 2.0 License. See the LICENSE file for details.