Paste a link
Every article and research page on this site has a markdown version at its address plus .md, such as https://commonsignals.org/methodology.md. A page whose address ends in a slash has index.html.md instead, as https://commonsignals.org/data/index.html.md does. Each one starts with the page's title, address, dates and, where the page has one, its citation line.
To hand over everything at once, paste https://commonsignals.org/llms.txt, which lists the studies, articles and methods pages with a line on each, or https://commonsignals.org/llms-full.txt, which holds them all in one file. Any assistant that can read a page you give it can use them.
Connect the server
Tools that support remote MCP servers, often called custom connectors, can connect to https://commonsignals.org/mcp. The server needs no login or key, runs no model and stores nothing. It is read-only: your tool can look up the studies, their figures, codebooks and coded comments, and the pages listed in llms.txt, and nothing else.
Add it wherever your tool keeps its connectors or MCP servers, with that address and no authentication. The steps differ from tool to tool and change often, so follow your tool's own guide:
- Claude: Get started with custom connectors using remote MCP
- ChatGPT: Connecting custom MCP servers in ChatGPT
- Cursor: Model Context Protocol (MCP)
Other tools that support remote MCP servers over Streamable HTTP can connect the same way. Each address can make 120 requests a minute.
The rules your tool receives
The server gives every tool that connects these instructions, word for word, and each tool's description repeats the rule that applies to it.
Common Signals codes public comment threads about AI to see how messages about AI risk are received, and publishes the coded data and the articles drawn from it. Follow these rules whenever you use these tools.
- Give the study and the base for every figure, rough ones such as "about half" included: say which study a number comes from and how many coded comments it is a share of. For stance, say whether the base is all coded comments or only those taking a clear position. Where a page or record gives a figure without its base, leave it out unless the tools give you the count.
- These are people who chose to comment on particular posts. They are not a sample of the public, of a country or of any demographic, so never present these figures as public opinion.
- Where a dimension has a reliability grade, follow it; the grade and its meaning come with each figure. Where a study has no blind reliability check, say its coding has been spot-checked only.
- Each study codes frames with its own list, so a frame in one study cannot be compared with a frame in another, even when the labels match. Never set two studies' frame figures side by side, even to warn against comparing them. Compare studies only on the shared dimensions that compare_studies returns.
- Common Signals describes how messages were received. It takes no side between communicators focused on present-day harms and those focused on existential risk, and it does not recommend one framing over another. If someone asks for advice, keep what these studies show separate from findings published elsewhere, and name the source of anything you bring in.
- The messenger map is a directory. Common Signals has not measured how far audiences trust anyone on it, so do not rank or recommend messengers from it.
- Quote comments exactly as returned, with their id and link, however short and even in passing. That includes a comment a page quotes: find it with search_comments first. Comment text belongs to its authors; the coding and compilation are licensed CC BY 4.0. Cite each study with the citation line its responses give, written out in full.
What the tools do
This list is generated from the server's own tool definitions.
list_studies
Lists every Common Signals study: slug, title, platform, the post it reacts to, capture dates, coded count, clear-position base, central claim, version, article and explorer links, and reliability status. Start here to find a slug for the other tools. These are people who chose to comment on particular posts, not a sample of the public, so never present their figures as public opinion.
It takes no parameters.
get_study
Returns one study's record (central claim, stance meanings, counts, capture method, coding check, changelog, licence, notes) and its figures: counts and percentages for every coded dimension, each with its base, and stance on two bases, all coded comments and only those taking a clear position. split_by repeats the figures for each group of one group dimension. Give the study and the base for every figure you use; figures_without_counts lists the record's figures that lack one.
| Parameter | Type | Required | What it is |
|---|---|---|---|
slug | string | yes | Study slug, from list_studies |
split_by | string | no | One of the study's group dimensions, such as period; get_codebook lists them |
get_codebook
Returns one study's codebook: every dimension it codes, with each value's label, definition and example, plus the central claim and what each stance value means. Check what a code means here before reporting it. Each study codes frames with its own list, so a frame in one study cannot be compared with a frame in another, even when the labels match.
| Parameter | Type | Required | What it is |
|---|---|---|---|
slug | string | yes | Study slug, from list_studies |
search_comments
Returns coded comments from one study, filtered by codes (dimension to allowed values, from get_codebook) and optional case-insensitive text, with n_matched and the coded count for the base. Up to 25 rows a call; page with offset. Quote a comment only whole and exactly as returned, curly quotes included, in a block quote, and name every comment, however short, by its id linked to its url. Comment text belongs to its authors; the coding is CC BY 4.0.
| Parameter | Type | Required | What it is |
|---|---|---|---|
slug | string | yes | Study slug, from list_studies |
filters | object of dimension to value or list of values | no | Dimension key to the value or values to keep, as in get_codebook; flags take true or false |
text | string | no | Case-insensitive text the comment must contain |
limit | whole number, 1 to 25 | no | Rows to return: 10 unless set, 25 at most |
offset | whole number, 0 or more | no | Matching rows to skip, for the next page |
compare_studies
Compares one shared dimension across every study: one entry per study with its own counts, percentages and base, and stance on both bases. Only dimensions that every codebook marks as shared are accepted. Frame is refused, because each study codes frames with its own list. Report each study's figure on its own base and never merge studies into one percentage.
| Parameter | Type | Required | What it is |
|---|---|---|---|
dimension | string | yes | A dimension every study marks as shared, such as stance or emotion |
list_pages
Lists the Common Signals pages indexed in /llms.txt: studies, articles, methods and data, and further reading, each with its section, title, page URL, markdown URL and description. Read one with get_page. When you use a page, name it as the source, and keep what these studies show separate from findings published elsewhere.
It takes no parameters.
get_page
Returns the markdown of one page listed by list_pages, given its page or markdown URL; other URLs are refused. Long pages come in parts: call again with part for the rest. Name the page as your source. Repeat its figures, rough ones included, only with the number of comments each is a share of; figures_without_counts lists those without one. Asked where to start reading, recommend only what the reading list lists. Quoted comment text belongs to its authors.
| Parameter | Type | Required | What it is |
|---|---|---|---|
url | string | yes | A page URL or markdown URL from list_pages |
part | whole number, 1 or more | no | Which part of a long page, from 1; each response says how many parts there are |
Licence and citation
Comment text belongs to its authors. The coding and compilation are licensed CC BY 4.0, and every study's coded data is on the data pages.
Every server response that carries figures or comments includes a cite line, built from the citation of the study's article and the address of its data page. Cite each study with that line. A markdown page carries its own citation line at the top, where the page has one.