conceptio-search
Resolves arXiv IDs and searches arXiv open-access preprints from the Conceptio archive.
Resolves DOI identifiers to their open-access document records in the Conceptio archive.
Searhes OWASP security standards and publications from the Conceptio archive.
Resolves PubMed IDs and searches PubMed/PMC biomedical literature from the Conceptio archive.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@conceptio-searchSearch for recent NIST publications on zero trust"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
conceptio-search
Search the open-access archive — papers, standards, textbooks, and legal documents — right from your terminal or your AI agent.
conceptio-search is a CLI and Model Context Protocol (MCP) server for the Conceptio Open Knowledge Archive. Every source in the archive is open access or public domain. The CLI authenticates with an API key: sign in once, save the key, search from anywhere.
For humans — search, export citations in 11 formats (BibTeX, APA, MLA, Chicago, IEEE, Harvard, RIS, Bluebook, OSCOLA, ISO 690, ANSI Z39), and download PDFs to disk with one command.
For AI agents — a stdio MCP server with eight tools, so Claude, Cursor, Windsurf, OpenCode, or any MCP client can search, queue batch searches, resolve identifiers (RFC, DOI, arXiv, PMID, PMCID, NIST/FIPS, W3C, US case citation), save PDFs into your workspace, and hand a document to Zotero or Obsidian.
100% self-contained — talks only to the public HTTPS API. No internal infrastructure; your key lives in
~/.conceptio/config.json.
Install
pip install conceptio-searchRequires Python 3.8+. (Checked, not just claimed: the suite parses every source file at the declared floor's grammar, so the promise cannot rot silently.)
The console script is conceptio; python -m conceptio_cli is the same entry
point for a virtualenv that was never activated.
Related MCP server: Crossref Academic MCP Server
Authenticate (one time)
The CLI needs an API key before it can search:
conceptio auth ckey_live_...Get the key by signing in at conceptio.app and
creating one on your profile (Free, Dev, Pro, or Enterprise). The key is stored in
~/.conceptio/config.json and sent as X-Api-Key with every request.
conceptio quota reports your tier at any time. Every command below
assumes this step is done — without a key the CLI refuses to run and tells
you exactly this.
Quick start
# Search — supports source:/lang:/category: directives
conceptio search "attention is all you need" --limit 5
conceptio search "source:nist zero trust" --license commercial-ok
conceptio search "source:eurlex AI act" --json
conceptio search "meditations marcus aurelius" --markdown # for Obsidian/Notion
conceptio search "diffusion models" --offset 20 # paginate past the first page
# Multi-query programmatic search — two modes:
conceptio search --batch queries.json --json # queue 1–50 searches, print the job handle
conceptio search --batch queries.json --wait --json # queue, then block until done and print results
conceptio search --batch queries.json --wait --timeout 60 # …but never block longer than a minute
conceptio search --batch queries.json --sync --json # run 1–10 immediately, no queue (one request)
conceptio search-job <job-id> --json # poll once; repeat on your own cadence
conceptio search-job <job-id> --wait --json # poll until done, then print results
# Resolve a known identifier straight to its document(s)
conceptio resolve "RFC 2119"
conceptio resolve "doi:10.1145/3290605.3300333"
conceptio resolve "2604.08499" # arXiv
conceptio resolve "PMID 41961061" # PubMed
conceptio resolve "PMC10601397" # PubMed Central
conceptio resolve "NIST FIPS 199"
conceptio resolve "410 U.S. 113" # US case citation (Supreme Court)
conceptio resolve "20-5364" # federal docket
conceptio resolve "RFC 2119" --json
# Download the original PDF
conceptio download 297465 -o paper.pdf # by document ID
conceptio download "https://arxiv.org/pdf/2604.21816v1.pdf" # by URL
# Citations — 11 formats: bibtex, apa, mla, chicago, ieee, harvard,
# ris, bluebook, oscola, iso690, ansiz39
conceptio cite 7288 --format bibtex
conceptio cite 7288 --format apa
conceptio cite 7288 --format iso690
# Document metadata
conceptio info 2844
conceptio proof 2844 # machine-readable evidence bundle
conceptio proof 2844 -q "quantum" # passage-level proof for a phrase
# License key (Pro) or API key + quota
conceptio auth CONCEPTIO-XXXX-XXXX-XXXX # Pro license (one-time, account-bound)
conceptio auth ckey_live_... # API key (agent/machine credential)
conceptio quota
conceptio quota --json # {tier, auth, credits…} for a status surface
# The flag may come before the subcommand, the way machine callers write it
conceptio --json search "zero trust"
# Version
conceptio --version
# AI agent server
conceptio mcpExample
$ conceptio search "source:nist zero trust" --limit 3
Conceptio - 3 results for "source:nist zero trust"
# Year Title Author Source PDF
1 n.d. NIST SP 1800-35: Implementing a Zero Trust Scott Rose (NIST); … NIST [x]
2 n.d. NIST SP 800-207: Zero Trust Architecture Scott Rose (NIST); … NIST [x]
3 n.d. NIST SP 800-207A: A Zero Trust Architecture Ramaswamy Chandramo… NIST [x]
Tip: conceptio download 2844 saves the PDF, conceptio cite 2844 exports a citation.Query words are highlighted in gold in the title column. The PDF column shows
[x] when a direct PDF link is available; [ ] rows still surface the source
URL via conceptio info <id>.
Search directives
The CLI understands the same directives as the Conceptio web app — they are stripped from the query and applied as real filters:
Directive | Example | Effect |
|
| Restrict to one or more sources ( |
|
| Restrict to a language (ISO code) |
|
| Restrict to a category tab |
|
| Keep only sources whose catalog metadata explicitly permits commercial use; this is a conservative filter, not a redistribution grant. |
Directives can be combined freely: conceptio search "cat:\"Computer Science & Tech\" source:nist lattice cryptography".
Resolve
conceptio resolve recognises several identifier shapes and returns a typed
response (kind: rfc | doi | arxiv | pmid | pmcid | nist | w3c | case plus
fallback null for plain text). Unrecognised identifiers degrade to a regular
search so the command never fails silently.
Model Context Protocol (MCP)
The conceptio mcp command starts a stdio JSON-RPC MCP server. It is dependency-free
(no MCP SDK required) and works with any MCP client.
Tools
Tool | Description |
| Search the archive ( |
| Run multiple searches in one call — queue 1–50 as a background job and return an opaque handle ( |
| Resolve a known identifier — RFC ( |
| Download the original open-access PDF for a Conceptio document ID or a direct PDF URL to a local path. |
| Get a citation for a document ID in any of 11 formats (BibTeX, APA, MLA, Chicago, IEEE, Harvard, RIS, Bluebook, OSCOLA, ISO 690, ANSI Z39). |
| Full metadata (title, author, source, category, license, year, language, URL, direct PDF URL, description) for a document ID. |
| Save one document to Zotero, or open a metadata-only Obsidian handoff. Server-side ownership, connector trial and failure semantics apply. |
| Bulk-save up to 500 document ids to Zotero. Requires Pro, institutional or licensed access — there is no free-trial path. |
Claude Desktop
Add to claude_desktop_config.json:
macOS:
~/Library/Application Support/Claude/claude_desktop_config.jsonWindows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"conceptio": {
"command": "conceptio",
"args": ["mcp"]
}
}
}Cursor
Settings → Cursor Settings → MCP → Add new MCP server:
{
"mcpServers": {
"conceptio": {
"command": "conceptio",
"args": ["mcp"]
}
}
}OpenCode / Windsurf / Antigravity
The same JSON shape works in any MCP-aware client — point it at conceptio mcp
and the eight tools above are exposed automatically. An MCP host may supply the
credential by environment (CONCEPTIO_API_KEY, or CONCEPTIO_BEARER_TOKEN for a
signed-in session) instead of writing ~/.conceptio/config.json.
After adding, restart the client and you can ask, for example:
"Find the NIST post-quantum encryption standard and download the PDF into my workspace."
Manual smoke test
echo '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | conceptio mcpWindows PowerShell quirk: pwsh pipes text to native commands as UTF-16, which the server (UTF-8 JSON-RPC) answers with a
Parse error. Real MCP hosts speak UTF-8 and are unaffected — this only bites hand-rolledecho ... | conceptio mcpprobes from pwsh. Save the request as UTF-8 first instead:'{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | Out-File -Encoding utf8 req.jsonl cmd /c "conceptio mcp < req.jsonl"
Fair use
Every source in Conceptio is open access or public domain. The CLI sends your API key with every request and is rate-limited per tier (see the rate limits); the browser keeps a separate free trial for casual searching without an account.
Add a credential once with conceptio auth <key> — it is stored in
~/.conceptio/config.json and sent with every request. conceptio auth
accepts either a self-hosted API key (ckey_live_..., the
agent/machine credential issued from your account — store it plainly, it is
hashed server-side and shown only once) or a Pro license key
(CONCEPTIO-XXXX-XXXX-XXXX, one-time / account-bound). API keys are
X-Api-Key; license keys are X-License-Key; exactly one credential is sent.
Without a saved key, every data command exits before touching the network:
Authentication required — save an API key before searching.conceptio quota reports your current tier (public / pro /
enterprise) and which auth the server honored (api_key / license /
firebase / public); conceptio quota --json returns the same report as
the structured object a status surface can render, with the human credential
hint on stderr so stdout stays parseable.
Configuration
State lives in ~/.conceptio/config.json:
{
"api_base": "https://www.conceptio.app",
"license_key": "",
"api_key": "",
"default_limit": 10,
"default_citation_format": "bibtex"
}Override the API base (for staging, self-hosted mirrors, or a local proxy) by
editing api_base. Public API origins must use HTTPS; only loopback HTTP
origins such as http://127.0.0.1:8000 are allowed for local development.
The client refuses cross-origin API redirects so credentials cannot be sent to
an unexpected host. PDF downloads reject loopback/private-network targets,
limit responses to 100 MiB, and write atomically. The MCP server additionally
keeps output_path beneath its current workspace and rejects traversal or
symlink escapes. Set default_limit to change the search page size; set
default_citation_format to any of the 11 supported format names.
Host processes (editors, CI, agent runtimes) can supply credentials without
writing to the config file — environment variables win over saved values:
CONCEPTIO_API_KEY, CONCEPTIO_LICENSE_KEY, and CONCEPTIO_API_BASE. A
signed-in human session can also be passed per-run as
CONCEPTIO_BEARER_TOKEN (the same short-lived bearer token the web app uses);
it takes precedence over a machine key, and exactly one credential is ever
sent.
Development
pip install -e ".[test]" # or: pip install -e . && pip install pytest
pytest tests/ # offline tests (mocked HTTP, no network)
python tests/live_check.py # live: the real CLI against a loopback stub
python tests/live_prod_check.py --yes # smoke: the real CLI against the real APIThe offline suite patches httpx with a MockTransport and stubs
ConceptioClient where it composes other services. No test in it makes a real
network call — which is also its ceiling: it can prove the client composes a
request and parses a response, but nothing about what the CLI decides from a
live exchange.
tests/live_check.py covers that gap. It starts the shared loopback stub
(conceptio-nvim/test/stub_api.py, dialled into done/running/expired job
modes and exhausted/pro/not_configured connector modes), points the real
binary at it with CONCEPTIO_API_BASE and an environment credential, and asserts
on exit codes, on stdout and on the stub's request log. It covers the whole
surface rather than a sample of it — search (directives, filters, markdown,
batch, jobs), resolve, cite, info, proof, download's boundary, the credential
paths, the connector failure reasons, all eight MCP tools, and the exact argv
each editor client sends (see below). It also refuses to run quietly against the
wrong binary: if conceptio --version does not match the version this tree
declares, every check would be proving an older install, so it fails and says so.
Everything stays on loopback: no account, no key, no production credit, and the
CLI is given a throwaway config directory so your real
~/.conceptio/config.json is never read or written. CONCEPTIO_CLI selects
the binary, CONCEPTIO_STUB the stub, CONCEPTIO_LIVE_VERBOSE=1 echoes each
command's output. It exits nonzero and prints every failing check with the
command's own output.
tests/live_prod_check.py is that harness's deliberate opposite, and closing
the pair is the point. Everything above is pinned against a server we control,
so none of it can see the live API moving — a renamed field in /api/me, a
refusal that changes wording, an edge rule that starts rejecting the CLI's
transport. Those ship silently, because this repo's own suite stays green while
every real user sees the drift.
It is opt-in (--yes) and safe by construction rather than by discipline: it
refuses to run unless the base is the shipped default, the credential is a
placeholder that any inherited CONCEPTIO_* variable is stripped out in favour
of, and it only runs reads the API keeps open without a credential plus commands
whose expected outcome is a refusal — no search ever executes. rm -f the
guard and it stops being able to tell production from a stub; re-point the
default base and 6 of its 9 checks go red, which is how you know they bite.
The client contract
The CLI is also a dependency of five editor clients — the Neovim plugin, the
VS Code extension, the Obsidian plugin, the Raycast extension and the Alfred
workflow. They exec this binary with an argv list and read stdout; nothing
couples them but that grammar. tests/client_contract.py records every
invocation they make, with the file it came from:
tests/test_cli_contract.pyparses each row againstcli.build_parser(), and while a client's checkout sits next to this repo, asserts its sources still contain the subcommand and flags the row claims.tests/live_check.pyruns the same rows against a real binary on the stub.
If you change a subcommand or a flag, one of the two will fail and name the client. Add a row when a client gains a call.
The README is a client too. tests/test_readme_contract.py extracts every
invocation from the code blocks above and parses it against the same grammar,
so a renamed flag cannot leave the documentation teaching a command that exits
with a usage error. Shell constructs (a pipe, a redirect) are skipped by an
explicit rule rather than by exception, and the test fails if it stops finding
the examples — an extractor that quietly matches nothing proves nothing.
License
MIT — see LICENSE.
Part of the Conceptio Open Knowledge Archive — 500+ living sources: arXiv, NIST, OWASP, CISA, PubMed/PMC, MIT OpenCourseWare, EUR-Lex, Project Gutenberg, and more.
This server cannot be deployed
Maintenance
Related MCP Connectors
Real-time Amazon, WIPO & PACER data for AI agents — 19 tools via the MCP protocol.
Free public MCP for AI agents — 193 tools, 44 workflows. No API key.
Search GitHub, npm, PyPI, StackOverflow, ArXiv from one MCP — built for coding agents.
Your org's AI agents, tasks, runs, search, and brain files as MCP tools and resources.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceEnables AI agents to download, index, and semantically search PDF research papers using 8 MCP tools.2GPL 3.0
- AlicenseAqualityDmaintenanceMCP server enabling AI agents to search and retrieve scientific papers, citations, and author profiles from Crossref, OpenAlex, and Semantic Scholar with no API keys required.518 PyPI3MIT
- AlicenseNot gradedqualityCmaintenanceEnables MCP agents to search academic literature across multiple scholarly engines, explore papers, authors, and references, inspect abstracts and full text, maintain saved paper collections, and export citations, BibTeX, abstracts, or full-text corpora.1Apache 2.0
- AlicenseNot gradedqualityBmaintenanceEnables MCP clients such as Claude to search over 450,000+ open-access papers, standards, and legal documents, download PDFs, retrieve citations, and fetch full document metadata.MIT