Your assistant asks whether the people you trust on UpTrust rely on one source, person or post, on a topic if you name one. It gets back whether it clears the bar you set (you would read it, share it, or stake your reputation on it) and a coarse band, never a raw score. The answer is your own network’s view, not a global rating. When your network has no record of it, the answer says so rather than calling it untrusted.
Ask it like this
- “Can I trust this newsletter on climate policy?”
- “Is this package safe to add, going by people I trust?”
When another tool fits better
- Which of these should I trust first?
You have several options and want them in order. - Whose judgment do the people I trust rely on for this topic?
You have no candidate yet and want to know whose judgment to read on a topic.
Before you use it
- It needs the Read your trust signal permission, which UpTrust asks you to approve when you connect.
- It only reads. It changes nothing on UpTrust.
- In Claude, connect
https://mcp.uptrusting.com/mcp. In ChatGPT, connecthttps://mcp.uptrusting.com/chatgpt. How to connect. - It can also answer from another trust graph than yours, such as one built from academic papers; see which communities it can answer from.
What your assistant is told
Your assistant decides when to use this tool from the title and description UpTrust’s server gives it. Here they are exactly as the server sends them.
- Tool name
is_trusted_by_principal- Title
- Can I trust this one source or person? (a go/no-go check)
- Description
Use this when the user asks whether one specific source, person or post is reliable — "can I trust this source?", "can I trust this person on X?", "is this site legit?", "can I cite this?", "is this safe to add?" — and as your own check on a single thing mid-task: before a source enters context, before a dependency is installed, before a paper is cited, before acting on a recommendation. The answer is how much the user's own trust network trusts it (the "principal" in the name is the signed-in user), or a named community's when you route the call to one, never a global rating: `trusted` says whether it clears the bar you set in `threshold` (`low` = would read it, `medium` = would share it, `high` = would stake reputation on it), alongside a 5-band verdict, never a raw score. For "trust them on X", pass X as `topic`. One entity at a time: use rank_for_principal to order several items the user already has, and top_trusted_for_principal to find trusted candidates in the first place. Pass this issuer's entity ID and its `entity_kind` (`user` or `post`) exactly as resolve_entity returned them; if the user handed you a URL or a name, call resolve_entity first. A post is checked through the trust in its author, on the post's own topic unless you pass `topic`. READ `evidence` and `topic_applied` before relaying the verdict: a band on `state: "none"` rests on nothing recorded (say that, never "low trust"), a band on `state: "thin"` rests on under one act of evidence, and `topic_applied: false` means the answer is unscoped no matter what topic you passed. Both fields are optional — an issuer that omits them is not reporting a basis, which is not the same as reporting that there is none. NEVER CARRY A BASIS ACROSS CALLS. The evidence reported for an unscoped check is the basis of THAT answer only; attaching it to a topic-scoped band from a second call produces a claim neither answer made ("high on immunology, on 47 acts"). When a topic-scoped answer reports no basis, the honest relay is that it reports none.