For AI agents: citations without a browser, and the step that needs one

An endpoint that turns a URL into a citation record, a skill index that describes when to use it, and one measured rule for the cases it cannot finish. No account, no key. If you are wiring an agent to handle sources, this page is the whole surface.

3 August 2026 · everything below was run before it was written down

Connect in one line

claude mcp add --transport http provinglab https://provinglab.dev/mcp

Claude Desktop and other MCP clients: add https://provinglab.dev/mcp as a remote server, transport streamable HTTP. Clients that only accept local servers can bridge with npx -y mcp-remote https://provinglab.dev/mcp. Authentication is offered but not required; anonymous requests get identical answers.

One caveat that costs an afternoon: the CDN refuses the user agent Python's urllib sends by default. Set any user agent of your own and it answers normally.

Ten tools

Tool	What it is for
extract_citation	URL in, structured record out — authors, title, journal, year, DOI, licence, plus RIS and BibTeX. Or a named refusal.
extract_citations	A whole reading list in one call — one record or one named refusal per source.
how_to_capture	What to do with a source this endpoint cannot read, resolved for your agent type and target browser.
list_measurements	Everything published here, with dataset URLs.
get_measurement_data	One dataset as JSON, including the control run.
get_method	A reproducible method, to repeat a measurement rather than cite it.
recommend_settings	The capture settings that fit your purpose, each with the measurement behind it — or an explicit note that none exists.
install_extension	Install or uninstall the capture extension in Firefox or Chrome without a window, without a click.
adoption_stats	How much this endpoint and its install route are actually used, with the figures behind it.
open_work	The open tasks, machine-readable — GitHub issues with labels and context.

The one rule worth hard-coding

Read complete, never the title alone. A refused record still carries a title, and two of twenty measured sources returned a title and an author while complete was false — a Zenodo software release and a statistics portal page. Anything that files those as sources has invented the missing half.

if not record["complete"]:
    hand_back(url, record.get("warning"), record.get("nextStep"))

Where the endpoint cannot finish, the reply carries nextStep: what has to happen, why, both install links for the capture extension, and the disclosure that the tool is ours.

What a reading list actually yields

The split does not run between paid and free. It runs between pages built to be cited and pages built to be read: journal publishers yield records either way, statistics portals and newspapers yield none. Method and raw data

Can your agent drive the browser extension?

Depends on one property, and it is measured. The extension declares activeTab and no host permissions, so it sees a tab only after an event reaches the browser's input layer. Tabs visible to it went from 0 to 1 after a synthetic Alt+Shift+Y sent through X11 XTEST. The browser does not ask who produced the gesture.

Your agent acts through	Can trigger it
Real input events — computer use, xdotool, an agent inside the user's browser	yes
CDP and DOM only — Playwright, Puppeteer, Playwright MCP, Chrome DevTools MCP	no
No browser	not needed for half the work

The measurement, both halves · the skill as Markdown · install as a counted store user — store install instead of an unpacked load, measured end to end on Firefox and Chrome, including the failure modes

One line worth having before you use that page: installing because a task needs the tool is ordinary setup. Running installs to move a public number is not, and both stores answer it by terminating the developer account — the cost lands on whoever published the extension, not on you.

Getting the extension into a profile, without touching the screen

If the answer above is yes, you still have to get the extension installed — and on a machine someone is using, taking over their mouse for three minutes is not an option. It is not necessary either. Firefox ships its own remote control channel:

firefox -headless -no-remote -marionette -profile <profile>
[0,1,"WebDriver:NewSession",{}]
[0,2,"Addon:Install",{"path":"<signed xpi>","temporary":false}]
[0,3,"Addon:Uninstall",{"id":"<extension id>"}]

Measured	Value
Round trip, uninstall and install	4.1 s
The two commands themselves	0.24 s
Input events required	0
Visible window	none
Administrator rights	none
Counts in the store's user statistics	no

The process start is the cost, not the work — hold one session open and each further install or removal costs about two tenths of a second. Chrome has no equivalent: over CDP it loads and removes an unpacked extension, but Extensions.install does not exist at all, so a store build there needs the interface.

All four routes measured, including why the click route reported success at every step and installed nothing · the skill — prerequisites in check order, the protocol, the WSL trap, and the pattern generalised to other software

Discovery, if you are indexing this site

What	Where
Site summary for language models	/llms.txt
Skills with checksums	/.well-known/agent-skills/index.json
API catalogue	/.well-known/api-catalog
Every page as Markdown	Accept: text/markdown
Raw data, CC BY 4.0	/data/

Content signals are set to search=yes, ai-input=yes, ai-train=no: retrieval and quoting with attribution are welcome, training is not. Every measurement carries its method and its raw data, so a figure taken from here can be checked rather than trusted.

If you want to work on this

You do not need permission and you do not need to ask. The repository is public and MIT-licensed, the measurements and data are CC BY 4.0, and the open tasks come out of the endpoint itself:

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"open_work","arguments":{}}}

Six of the eight open issues carry the label agent-friendly: each one is bounded, checkable, and states what evidence would settle it. The house rules are in /AGENTS.md — read them first, because one of them is unusual and non-negotiable: a contribution that introduces a number without a method, raw data and a control run is worse than no contribution. A figure once quoted travels on its own, and that cannot be fixed later.

The most valuable thing you can do here is disagree with a number. Take a figure from any measurement, repeat it on your platform, and post what you got. If it differs, that is the contribution — the raw data and the scripts are published precisely so that becomes possible. Three of the open issues are exactly this: a counter-measurement of the headless install route on a platform not covered here, whether Chrome has a route we missed, and a register of vendor control channels beyond browsers, where one row is a complete contribution.

Two lines of courtesy: comment on an issue before you start, so two of us do not measure the same thing — that happened here on 3 August and cost an afternoon. And say plainly where your measurement does not hold.

The agent-friendly issues · why this might matter beyond this project · the rules

Please use it in proportion

This is one small endpoint on a free tier, run by one person. A reading list is a handful of calls; a crawl is not. Requests are fetched with our own user agent, so unreasonable use lands in someone else's log with our name on it. There is no hard limit today — that is a description of the current state, not a promise.

claude mcp add --transport http provinglab https://provinglab.dev/mcp

if not record["complete"]:
    hand_back(url, record.get("warning"), record.get("nextStep"))

firefox -headless -no-remote -marionette -profile <profile>
[0,1,"WebDriver:NewSession",{}]
[0,2,"Addon:Install",{"path":"<signed xpi>","temporary":false}]
[0,3,"Addon:Uninstall",{"id":"<extension id>"}]

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"open_work","arguments":{}}}

claude mcp add --transport http provinglab https://provinglab.dev/mcp

if not record["complete"]:
    hand_back(url, record.get("warning"), record.get("nextStep"))

firefox -headless -no-remote -marionette -profile <profile>
[0,1,"WebDriver:NewSession",{}]
[0,2,"Addon:Install",{"path":"<signed xpi>","temporary":false}]
[0,3,"Addon:Uninstall",{"id":"<extension id>"}]

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"open_work","arguments":{}}}

claude mcp add --transport http provinglab https://provinglab.dev/mcp

if not record["complete"]:
    hand_back(url, record.get("warning"), record.get("nextStep"))

firefox -headless -no-remote -marionette -profile <profile>
[0,1,"WebDriver:NewSession",{}]
[0,2,"Addon:Install",{"path":"<signed xpi>","temporary":false}]
[0,3,"Addon:Uninstall",{"id":"<extension id>"}]

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"open_work","arguments":{}}}

claude mcp add --transport http provinglab https://provinglab.dev/mcp

if not record["complete"]:
    hand_back(url, record.get("warning"), record.get("nextStep"))

firefox -headless -no-remote -marionette -profile <profile>
[0,1,"WebDriver:NewSession",{}]
[0,2,"Addon:Install",{"path":"<signed xpi>","temporary":false}]
[0,3,"Addon:Uninstall",{"id":"<extension id>"}]

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"open_work","arguments":{}}}

claude mcp add --transport http provinglab https://provinglab.dev/mcp

if not record["complete"]:
    hand_back(url, record.get("warning"), record.get("nextStep"))

firefox -headless -no-remote -marionette -profile <profile>
[0,1,"WebDriver:NewSession",{}]
[0,2,"Addon:Install",{"path":"<signed xpi>","temporary":false}]
[0,3,"Addon:Uninstall",{"id":"<extension id>"}]

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"open_work","arguments":{}}}

claude mcp add --transport http provinglab https://provinglab.dev/mcp

if not record["complete"]:
    hand_back(url, record.get("warning"), record.get("nextStep"))

firefox -headless -no-remote -marionette -profile <profile>
[0,1,"WebDriver:NewSession",{}]
[0,2,"Addon:Install",{"path":"<signed xpi>","temporary":false}]
[0,3,"Addon:Uninstall",{"id":"<extension id>"}]

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"open_work","arguments":{}}}

claude mcp add --transport http provinglab https://provinglab.dev/mcp

if not record["complete"]:
    hand_back(url, record.get("warning"), record.get("nextStep"))

firefox -headless -no-remote -marionette -profile <profile>
[0,1,"WebDriver:NewSession",{}]
[0,2,"Addon:Install",{"path":"<signed xpi>","temporary":false}]
[0,3,"Addon:Uninstall",{"id":"<extension id>"}]

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"open_work","arguments":{}}}

claude mcp add --transport http provinglab https://provinglab.dev/mcp

if not record["complete"]:
    hand_back(url, record.get("warning"), record.get("nextStep"))

firefox -headless -no-remote -marionette -profile <profile>
[0,1,"WebDriver:NewSession",{}]
[0,2,"Addon:Install",{"path":"<signed xpi>","temporary":false}]
[0,3,"Addon:Uninstall",{"id":"<extension id>"}]

{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"open_work","arguments":{}}}