
MCP server for open problems with exact checkers: Quantavane's benchmark, verified results, replays
MCP server for open problems with exact checkers: Quantavane's benchmark, verified results, replays and citations
·
An MCP server for open problems with exact checkers. It gives an AI agent, or a person at a terminal, Quantavane's Open Problems Benchmark v2.1: 149,372 open problems, each with its best known value, the source and date of that value, and an exact checker. 395 are single problems (Tao's optimization constants, Ramsey numbers, cages, kissing numbers, sorting networks, Hadamard matrices, matrix multiplication and more); 148,977 are cells of six families read from their keepers' tables (covering designs, spherical codes, linear codes, covering arrays, constant weight codes and project scheduling). Two collections of our own come with them: Quantavane Prestige 100, one hundred famous open problems (R(5,5), the kissing numbers from dimension 5 to 31, 3 x 3 matrix multiplication, the Moore graph of degree 57 and more) with their keepers' records, and the Quantavane Originals, 39 new problems of ours, each with its own checker, a written novelty check and our first value. A candidate is scored by the benchmark's own checker, in a sandbox, offline, and the answer is a verdict and an exact value, never an estimate.
run_claim_checks runs Quantavane's claim pipeline on an object: two separately written checkers must agree on its
exact value, red controls (copies of it broken so that they must fail) must be refused, and the value must beat the
keeper's value on file. It also answers from Quantavane's verified results (the open dataset), our independent replays
of certificates in the optimization constants repository, and gives citations for all of it.
You need uv; it fetches Python 3.10 or later if your system lacks it. The first command downloads the server from GitHub (about 15 seconds); the rest take a second or two each.
# 1. Check the install: every bundled file against its SHA-256, then reference certificates through the checkers
uvx --from git+https://github.com/arhancanli/quantavane-mcp quantavane-mcp selftest
# 2. Check a covering design C(17,7,4,6) with 17 blocks (ours, listed by the Covering Repository)
curl -sSLO https://raw.githubusercontent.com/arhancanli/quantavane-benchmark/main/refs/ours_covering_17-7-4-6.txt
uvx --from git+https://github.com/arhancanli/quantavane-mcp quantavane-mcp check covering-17-7-4-6 ours_covering_17-7-4-6.txt
# "verdict": "certified", "value": "17", "standing": "equals the best known"
# 3. Remove one block and check again: refused, with the 6-subset that is no longer covered
sed 1d ours_covering_17-7-4-6.txt > broken.txt
uvx --from git+https://github.com/arhancanli/quantavane-mcp quantavane-mcp check covering-17-7-4-6 broken.txt || true
# "verdict": "not_certified", "reason": "the 6-subset [2, 3, 5, 8, 9, 11] meets every block in fewer than 4 points"Then give the tools to an agent, for example Claude Code (other clients: connect a client):
claude mcp add quantavane -- uvx --from git+https://github.com/arhancanli/quantavane-mcp quantavane-mcpand ask it to pick an open problem, build candidates in code and check them. check and claim print JSON, so they
also work in any scoring pipeline without an MCP client.
Python 3.10 or later on Linux or macOS. The Ramsey checker is C, so Ramsey problems also need a C compiler (cc).
uvx quantavane-mcp --version # run without installing (uv)
pip install quantavane-mcp # or install with pip
quantavane-mcp selftest # checks the bundled data and runs reference certificatesFrom the source: uvx --from git+https://github.com/arhancanli/quantavane-mcp quantavane-mcp, or in a clone,
pip install .. A client can start the server from GitHub the same way: wherever this README says uvx quantavane-mcp, write uvx --from git+https://github.com/arhancanli/quantavane-mcp quantavane-mcp (in JSON,
"args": ["--from", "git+https://github.com/arhancanli/quantavane-mcp", "quantavane-mcp"]).
selftest verifies every bundled file against its SHA-256, then runs a covering design, a constant weight code, a
spherical code, a constant, a Ramsey colouring and one of the Originals through the sandboxed checkers, compares the
verdicts with the benchmark's, and runs the claim checks once.
The server speaks MCP over stdio. Any client starts it with the command uvx and the argument quantavane-mcp.
Claude Code
claude mcp add quantavane -- uvx quantavane-mcpor, for a project, in .mcp.json:
{
"mcpServers": {
"quantavane": { "command": "uvx", "args": ["quantavane-mcp"] }
}
}Claude Desktop (claude_desktop_config.json; on macOS in ~/Library/Application Support/Claude/, on Windows in
%APPDATA%\Claude\). Desktop apps do not always see your shell's PATH, so give the full path of uvx
(which uvx). They may also start servers in the root folder, so name the folders your certificate files live in:
{
"mcpServers": {
"quantavane": {
"command": "/full/path/to/uvx",
"args": ["quantavane-mcp"],
"env": { "QUANTAVANE_MCP_ROOTS": "/Users/you/candidates" }
}
}
}Cursor (~/.cursor/mcp.json) uses the same mcpServers block as Claude Desktop.
VS Code (.vscode/mcp.json):
{
"servers": {
"quantavane": { "type": "stdio", "command": "uvx", "args": ["quantavane-mcp"] }
}
}Codex CLI (~/.codex/config.toml):
[mcp_servers.quantavane]
command = "uvx"
args = ["quantavane-mcp"]Other clients: command uvx, arguments ["quantavane-mcp"], transport stdio. With a pip install the command
is quantavane-mcp (or python -m quantavane_mcp).
| Tool | What it does |
|---|---|
list_problems | The problems, one line each, with pages. Without a family it lists the single problems; with family it lists that family (its single problems, then its cells), filtered by the cells' parameters (params, such as {"v": 13, "t": 3}) or by words, or draws a random sample of cells with a seed (the same draw as obb.py sample). |
get_problem | One problem or family cell (a cell is written out like a problem file, as obb.py show does): statement, objective, best known value with source, keeper and read date, values in review, certificate format, checker, improvement rule, reference certificates, a cell's parameters and the table's other bound. |
check_candidate | Runs the problem's exact checker on a certificate (text, or a local file) in the sandbox. Verdict, exact value, comparisons with the best known values, the certificate's SHA-256 and the commands that rerun the check with the public benchmark. |
compare_to_record | Is a value better than the best known? Exact comparison with the page best and the verified values in review, with the date the value was read. Proves nothing about any object. |
search_results | Quantavane's verified results by table, cell (CAN(3,2872,4), C(84,7,2,5), A(51,6,6), R(8,12), [63,14,33] over GF(5), ...), words, date or status, with citation info. |
get_replay | Our replay of a certificate in teorth/optimizationproblems, by pull request, row (1c Kos2026) or constant: what was checked, the exact values, our programs and the commands that re-derive it. |
cite | BibTeX and citation text (and CITATION.cff) for the benchmark, a problem, a result, the dataset, a replay or this server. |
list_collection | Prestige 100 (slot, problem, family, the record as its keeper lists it and who holds it) or the Originals (field, title, our best known value, status), with words and a family or field to filter by. |
get_collection_problem | One problem of a collection in full, by id or Prestige 100 slot: for Prestige 100 the record and its holders, the other bound, what would beat it, the keeper read with date and version, the history of the bounds, why it matters and our replayed certificates; for the Originals the statement, context, novelty check, exact first values and references. |
run_claim_checks | Quantavane's claim pipeline on one object: the benchmark's checker and a second checker written apart from it (both sandboxed) must agree on the exact value, red controls built from the object must be refused by both, and the value must strictly beat the keeper's value on file. RECORD, NOT-A-RECORD or NOT VERIFIED, with every check, the keeper's value and its date, the literature queries to run and the pipeline's packet as text. |
Every tool is read-only, has an input schema with limits and an output schema, and returns structured content.
Errors come back as tool errors that say what to change. get_problem, check_candidate, compare_to_record,
run_claim_checks and cite take the ids of the Originals (such as qr-sumfree) as well; the Prestige 100 problems are
single problems of the benchmark and keep their ids (such as ramsey-5-5).
Resources: quantavane://about (version, data commits and dates, limits), quantavane://problems and
quantavane://problems/{problem_id} (a single problem or a cell), quantavane://families and
quantavane://families/{name} (each family's keeper, date read, open rule and what was left out),
quantavane://checkers, quantavane://checkers/{program} (documentation) and
quantavane://checkers/{program}/source, quantavane://refs/{filename} (our reference certificates),
quantavane://benchmark/readme, quantavane://benchmark/changelog, quantavane://open-data/readme,
quantavane://collections, quantavane://collections/{name} (a collection's README),
quantavane://collections/{name}/{problem_id} (a problem's page, by id or Prestige 100 slot) and
quantavane://collections/{name}/refs/{filename} (the Originals' certificates). Problem ids, checker names, collection
names and certificate names complete as you type in clients that support completion.
Prompts: attack_problem(problem_id), verify_certificate(problem_id, certificate_path), explain_replay(pr),
prepare_claim(problem_id, certificate_path).
Outputs below are abridged from real calls.
Check a covering design from a file:
check_candidate {"problem_id": "covering-17-7-4-6", "certificate_path": "/tmp/ours_covering_17-7-4-6.txt"}
{
"verdict": "certified",
"value": "17",
"improves_page_best": false,
"comparisons": [{"reference": "page best", "value": "17", "threshold": "17", "improves": false,
"source": "Covering Repository (coveringrepository.com), C(17,7,4,6) = 17 blocks: ours, listed since 2026-10-26 (before it 18); read 6 Nov 2026"}],
"certificate_sha256": "cf4b0b58b318634eb94e629075d9a916d6aa5585c240bf75d1221e0cbb8d688d",
"reproduce": "git clone https://github.com/arhancanli/quantavane-benchmark && cd quantavane-benchmark && git checkout 9cb6183\npython3 -I obb.py check covering-17-7-4-6 CANDIDATE ..."
}Remove one block and the same call returns "verdict": "not_certified" with the checker's reason
(some 6-subset meets every block in fewer than 4 points).
A family cell: the open entries of Grassl's tables for binary codes of length 32, then one in full:
list_problems {"family": "linear-code", "params": {"q": 2, "n": 32}}
{"total": 2, "problems": [{"id": "code-q2-n32-k14", "title": "Linear code [32,14] over GF(2)", "best_known": "8"},
{"id": "code-q2-n32-k18", "title": "Linear code [32,18] over GF(2)", "best_known": "6"}]}
get_problem {"problem_id": "code-q2-n32-k14"}
{"params": {"q": 2, "n": 32, "k": 14}, "objective": "maximize",
"best_known": {"value": "8", "exact": "8", "read": "read 2026-11-09"},
"bound": {"value": "9", "kind": "upper bound", "source": "the upper bound in the same table"},
"provenance": {"keeper": "M. Grassl, bounds on the minimum distance of linear codes (codetables.de)",
"url": "https://www.codetables.de/BKLC/BKLC.php?q=2&n=32&k=14", "read": "2026-11-09", ...},
"checker": {"program": "lincode_check.py", "standalone": "python3 -I checkers/lincode_check.py 2 32 14 FILE", ...}}A [32,14] binary code with minimum distance 9 would close this entry. {"family": "linear-code", "sample": 5, "seed": 0} draws five cells at random, the same five that obb.py sample linear-code 5 0 prints.
Would 0.41035 be a new lower bound for the autocorrelation constant C_1c?
compare_to_record {"problem_id": "1c-lower", "value": "0.41035"}
{"value": "8207/20000", "improves_page_best": true, "improves_verified_in_review": false,
"best_known_read": "as on the page at main 87b4ebb; problem file of 2026-11-05"}It beats the page's 0.4103333 but not the certificate in review (0.41035965...), and it is only a number until a certificate for it is checked.
A covering array cell:
search_results {"cell": "CAN(3,2872,4)"}
{"covering_array_entries": [{"t": 3, "k": 2872, "v": 4, "N": 604, "previous_N": 607, "family_rows_k": 2876,
"source": "ca-tools (github.com/chrispbsmolen/ca-tools), VERIFIED"}], ...}Our replay of a pull request, with the commands to re-derive it:
get_replay {"pr": 292}
{"found": true, "matches": [{"heading": "C1a, C1c and C1d (3 to 7 November 2026)",
"rows": ["| $C_{1c} \\geq$ | [#292](https://github.com/arhancanli/quantavane-mcp/blob/main/...) (Koshta's row) | ... | 134648548624804778248790016 / 328144294688458468928334649 = 0.41033335274849333953... | ..."],
"rederive": ["git clone https://github.com/arhancanli/quantavane-replays.git",
"cd quantavane-replays && git checkout b892ccb", "./fetch.sh && ./run_all.sh"]}], ...}A Prestige 100 problem and one of the Originals:
list_collection {"collection": "prestige100", "group": "Ramsey", "limit": 2}
{"total": 24, "entries": [{"id": "ramsey-5-5", "slot": 1, "family": "Ramsey numbers", "record": "R(5,5) ≥ 43",
"holders": "G. Exoo (1989)", "flagship": true, "quantavane": false}, ...]}
get_collection_problem {"collection": "originals", "problem": "qr-sumfree"}
{"title": "Largest sum-free set of quadratic residues", "entry": {"field": "Number theory", "objective": "maximize",
"best_known": {"value": "174", ...}, "novelty": {"verdict": "new as stated. ...", ...}, "first_values": {...}}, ...}The claim pipeline on our 6-regular graph of girth 9 on 5,600 vertices:
run_claim_checks {"problem_id": "cage-6-9", "certificate_path": "/tmp/ours_cage_6_9_n5600.s6"}
{"verdict": "RECORD (5600 vs listed 7781; both checks OK, red controls all rejected; keeper value on file, read 8 Nov 2026; problem file of 2026-11-08; re-read the keeper today before sending)",
"verified": true, "record": true, "second_checker": "second_check.py cage", "red_controls": 1,
"checks": [{"kind": "check", "program": "cage_check.py", "verdict": "certified", "value": "5600"},
{"kind": "check", "program": "second_check.py cage", "verdict": "certified", "value": "5600"},
{"kind": "red", "name": "the edge 0 800 removed (two vertices lose a neighbour)", "program": "cage_check.py", "verdict": "not_certified"},
{"kind": "red", "name": "the edge 0 800 removed (two vertices lose a neighbour)", "program": "second_check.py cage", "verdict": "not_certified"}],
"keeper": {"name": "House of Graphs cage table (...)", "value": "7781", "read": "read 8 Nov 2026; problem file of 2026-11-08",
"re_read": "not done: offline. The pipeline re-reads the keeper's page the same day before a claim is sent; ..."},
"packet": {"VERDICT": "...", "checks.txt": "...", "keeper.txt": "...", "literature.json": "...", "README.md": "..."}}Without an MCP client, the same check from the command line (JSON out; exit status 0 certified, 1 not certified, 2 no verdict or an error), and the claim checks (exit status 0 RECORD, 1 NOT-A-RECORD, 2 NOT VERIFIED or an error):
quantavane-mcp check covering-17-7-4-6 design.txt
quantavane-mcp claim covering-17-7-4-6 design.txtattack_problem prompt, or give it the tools and the server's instructions. The loop is:
get_problem, build candidates in code, write each to a file, check_candidate with certificate_path, keep the
certified ones. Inline certificates are accepted up to 4,000,000 characters, but files keep the context small.verdict as the only ground truth: certified proves value; not_certified means the certificate is
invalid (reason says why); no_verdict means a limit stopped the check and nothing was proved either way.timeout_s goes up to 900
by default.total, has_more, next_offset); outputs are small, and large texts (certificates, checker
sources) are resources the agent reads only when it needs them.run_claim_checks (or the prepare_claim prompt): it adds the second checker and
the red controls, and says RECORD only when both checkers agree, every red control is refused and the value beats
the keeper's value on file.quantavane_mcp.checking.Checker class is the same sandboxed checker as a Python API
(Checker(store, settings).check_data(problem_id, data)), and quantavane-mcp check is a scoring command for any
pipeline. Gym environments around the same checkers can be attached later: set QUANTAVANE_MCP_GYM to a backend
(package.module:attribute) with environments() and propose(env_id, obj) and the server adds gym_environments
and gym_propose. Unset, it adds nothing and imports nothing.quantavane://checkers/{program} gives the checker's own
documentation (what it proves and how) and its source. The checkers are standard-library Python and one C program,
written apart from the searches that produced the reference results. They are the public benchmark's files at the
commit in quantavane://about, verified against their SHA-256 before every run.RESULT. exactness and
score_meaning in each result say which.obb.py for every certificate in its test.sh (the routing ones when the CVRPLIB files are
present).cite(target="problem") gives the reference.run_claim_checks is the gate Quantavane uses before it sends a result to a keeper. The second
checkers (quantavane_mcp/independent/second_check.py, standard library, runnable on their own) were written from
the problem statements, apart from the benchmark's checkers, for 27 checker programs: Ramsey and multicolour Ramsey,
Schur and weak Schur, van der Waerden, covering designs, constant weight codes, linear codes (over GF(2) to GF(9) by
listing the code), covering arrays, spherical codes (the first checker's printed largest cosine confirmed exactly to
one unit of its 20th decimal), kissing configurations, Golomb rulers, sorting networks, cap sets, no-three-in-line,
strong powers of cycles, Hadamard and circulant weighing matrices, Costas arrays, mutually orthogonal Latin squares,
projective planes, strongly regular graphs, cages, degree-diameter graphs, snakes and coils, superpermutations, trees
for Erdős problem #993 and matrix multiplication algorithms. Each red control breaks the object in a way that must
be refused (a clique planted, a block or an edge removed, a word or vector repeated, a point duplicated). The
analytic constants, routing, scheduling, the Mona Lisa tour, Erdős #107 and the Originals have no second checker yet,
so their claims come back NOT VERIFIED. A second checker that would exceed its limits says so (no verdict) rather
than guessing: the largest of our own cages (17,600 vertices), very large covering designs and codes, and sorting
networks on more than about 35 channels are past them.not_certified and a reason.
The Ramsey and covering checkers make good first exercises (small inputs, clear failures).QUANTAVANE_MCP_ROOTS to the folder students write their certificates to.quantavane://about (benchmark v2.1 with Prestige 100 and
the Originals, 504492c of 9 Nov 2026, its tables read on 8 and 9 Nov; open data of 7 Nov 2026; replays 57988b0 of 9
Nov 2026). Best known values move: check the source page before claiming a record (run_claim_checks compares with
the value on file and says so). A newer release of this package brings newer data.fetch.sh in a clone of the
benchmark for CVRPLIB and the Mona Lisa instance, unpack the scheduling sets as the family's instances note says,
and set QUANTAVANE_MCP_INSTANCES to the instances/ folder. CVRPLIB files and the Mona Lisa instance are checked
against the SHA-256 the benchmark pins before they are used. The server never downloads them. (The Originals'
instance files are ours and bundled.)run_claim_checks gives each program run 600 s by default (its second checkers are plain Python: a Ramsey
colouring on 282 vertices takes about three minutes there). Most checks take under a second; the Turán (42a) and Bohr radius (59a) checkers can take a minute or more
on large certificates, a linear code at the edge of its family about 20 s.QUANTAVANE_MCP_ALLOW_UNCAPPED=1; WSL works as Linux.All optional, as environment variables:
| Variable | Default | Meaning |
|---|---|---|
QUANTAVANE_MCP_ROOTS | working folder (unless / or home) and the temporary folders | Folders certificate_path may point into, separated by : (; on Windows). |
QUANTAVANE_MCP_TIMEOUT_S | 120 | Default time limit of one check, in seconds. |
QUANTAVANE_MCP_MAX_TIMEOUT_S | 900 | Largest timeout_s a caller may ask for. |
QUANTAVANE_MCP_MEMORY_MB | 1024 | Memory cap of one checker process. |
QUANTAVANE_MCP_MAX_BYTES | 33554432 | Largest certificate file. |
QUANTAVANE_MCP_MAX_CONCURRENT | 2 | Checks running at once. |
QUANTAVANE_MCP_INSTANCES | none | Folder with CVRPLIB instance files. |
QUANTAVANE_MCP_BENCHMARK_DIR | bundled copy | A local benchmark checkout to use instead (newer problems; not verified against the release). |
QUANTAVANE_MCP_CACHE | ~/Library/Caches/quantavane-mcp or ~/.cache/quantavane-mcp | Where the compiled Ramsey checker is kept. |
QUANTAVANE_MCP_CC | cc, then gcc, then clang | C compiler. |
QUANTAVANE_MCP_GYM | none | A training-gym backend, package.module:attribute. |
QUANTAVANE_MCP_ALLOW_UNCAPPED | unset | 1 runs checks where the memory cap cannot be enforced. |
certificate_path must resolve, after symlinks, to a regular file inside the allowed folders; hidden files and
folders are refused. The file is copied before the check, so the checker never sees your path.uv venv && uv pip install -e ".[test]"
pytest # the whole suite
QUANTAVANE_MCP_TEST_BENCHMARK=~/quantavane-benchmark pytest tests/test_parity_obb.py # parity with obb.py
QUANTAVANE_MCP_TEST_INSTANCES=~/quantavane-benchmark/instances pytest -k routing # routing references
python3 -I scripts/sync_data.py --open-data DIR # rebuild the bundled data from the sources at pinned commits
python3 -I scripts/sync_data.py --open-data DIR --check # confirm the bundled data matches the sourcesCode: MIT (LICENSE). The bundled data keeps the licence its source states (LICENSE-DATA.md): the benchmark's files (problems, families, checkers, reference certificates, the Originals) and the replay index are MIT, as the benchmark and the replays repository state; the Prestige 100 collection's data is CC BY 4.0 with its checkers MIT, as its index states; the verified results are CC BY 4.0, as the dataset states. The families hold values read from other people's tables, credited to their keepers and holders (LICENSE-DATA.md names each table); no keeper's file is included.
cite(target="server"), or:
@software{quantavane-mcp,
author = {Canli, Arhan},
title = {quantavane-mcp: an MCP server for open problems with exact checkers},
year = {2026},
version = {0.2.0},
url = {https://github.com/arhancanli/quantavane-mcp}
}To cite the benchmark, a problem, a result, the dataset or a replay, use cite with that target; citing a problem
also gives the source of its best known value, which should be cited as well. See also CITATION.cff.
Quantavane: https://quantavane.com. Questions and corrections: open an issue.
Pick your client and paste the snippet. Each one is the same server, written the way that client expects it.
claude mcp add quantavane -- uvx quantavane-mcp{
"mcpServers": {
"quantavane": {
"args": [
"quantavane-mcp"
],
"command": "uvx"
}
}
}code --add-mcp '{"name":"quantavane","args":["quantavane-mcp"],"command":"uvx"}'[mcp_servers.quantavane]
command = "uvx"
args = ["quantavane-mcp"]Runs locally on your device. Your client starts the server itself, so nothing has to be hosted.
Paste this prompt into your agent. It reads this page and does the setup for you.
Read https://aiagentslisting.com/mcp/quantavane-mcp to learn what the "Quantavane Mcp" MCP server does and how to install it. Add it to my coding agent's MCP configuration as documented on that page, then confirm the server connects and list the tools it exposes.Agents can also browse this directory over MCP at https://aiagentslisting.com/api/mcp
claude mcp add quantavane -- uvx --from git+https://github.com/arhancanli/quantavane-mcp quantavane-mcp
claude mcp add quantavane -- uvx quantavane-mcpuvx quantavane-mcp --version # run without installing (uv)
uvx --from git+https://github.com/arhancanli/quantavane-mcp quantavane-mcp selftest
uvx --from git+https://github.com/arhancanli/quantavane-mcp quantavane-mcp check covering-17-7-4-6 ours_covering_17-7-4-6.txt
uvx --from git+https://github.com/arhancanli/quantavane-mcp quantavane-mcp check covering-17-7-4-6 broken.txt || trueThis server runs locally, so we can't read its tool list over the web yet.
We could not run this server: its endpoint did not list its tools to an anonymous client, which usually means it asks for an API key or a sign-in. The four checks that need a live server are left out, and the grade tops out at B until we can read its tools.
Maintain this server? Claiming proves you own the listing and earns the last 4 points.
Claim this listingPaste this line near the top of your repository's README. The badge always renders the current score.
[](https://aiagentslisting.com/mcp/quantavane-mcp)Nothing comparable is listed yet.
No tagged releases on record.
Connect to the AI Agents Listing MCP server at https://aiagentslisting.com/api/mcp using the HTTP transport and complete the sign-in when it asks. Then call the submit_review tool for the listing "quantavane-mcp" with a rating from 1 to 5 and a short note on what worked and what did not.No reviews yet
Be the first to share your experience with this tool.
One email a week. New agents, MCP servers and skills, and what is actually getting traction.