The topics in a set of documents in one call, with each document's topic mixture. Nothing stored.
Scored across 2 tools
find_topics discovers topics from a set of documents and returns document-topic mixtures, while topic_of_document assigns a single document to the served model's existing topic set. The purposes and input constraints are explicit, so an agent should not confuse them.
Both tool names use clear snake_case and are descriptive. The only minor inconsistency is grammatical form: find_topics is verb-led, while topic_of_document is noun-led.
The server exposes only two tools, which is thin for a topic-modeling service. Both tools are core and non-redundant, but the surface is below the typical well-scoped range.
The core lifecycle is covered: discover topics from a corpus and place a new document on the general topic set. Minor gaps remain, such as batch assignment to the served topics or a direct way to inspect the served model's topic list.
Tuatara Topics: find the topics in a set of documents. Returns num_topics topics, each with its words, a label, a description and a weight, plus every document's topic mixture, in one call; a dictionary model's word knowledge guides a probabilistic topic model, so topics are coherent on small sets and short texts. Nothing is trained on or kept from the request. 4 to 500 documents of up to 20,000 characters per call.
| Name | Required | Description | Default |
|---|---|---|---|
| labels | No | give each topic a short label and description (default true) | |
| documents | Yes | the documents (plain text), 4 to 500 | |
| num_words | No | words per topic, 3-30 (default 15) | |
| num_topics | No | topics to find, 2-100 and at most half the number of documents (default 10) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnly, idempotent, non-destructive, closed-world, so the safety profile is covered. The description adds genuinely new behavioral facts: nothing is trained on or retained from the request, everything returns in one call, and there are per-call input limits. It does not discuss cost, latency, or determinism of results across identical calls, which would have pushed it higher.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Front-loaded with the verb and resource, then results, then behavior, then limits. Every sentence carries information, though the middle clause about the dictionary model guiding a probabilistic model is dense and slightly technical for an invocation decision.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no output schema, the description does the work of describing the return payload and its fields, plus input constraints, which is enough to call it correctly. Minor gaps are error behavior (what happens below 4 documents) and expected runtime for 500 documents.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so baseline is 3, but the description adds meaning beyond the schema: the 20,000-character per-document cap is not in the schema, and it explains the semantics of the num_topics parameter by tying it to the returned structure. It does not clarify num_words or the labels flag beyond what the schema already says.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb and resource ('find the topics in a set of documents') and immediately specifies the output shape (num_topics topics with words, label, description, weight, plus per-document mixtures). This is clearly distinguishable from neighbors like summarize_document or search, which do not extract latent topics.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Implies the intended domain ('coherent on small sets and short texts') and gives hard input bounds (4-500 documents, 20,000 chars each), which helps an agent decide feasibility. However, it never names an alternative tool or states when not to use this (e.g., single-document summarization should go to summarize_document).
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Tuatara Topics: place one document on the general topic set (the served model's topics) and return the topics it belongs to with their weights. Nothing is stored.
| Name | Required | Description | Default |
|---|---|---|---|
| document | Yes | the document (plain text) | |
| num_words | No | words per topic, 3-30 (default 15) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description discloses useful behavior beyond the annotations — it ranks a single document against the served model's topic set, returns weighted topics, and asserts "Nothing is stored." However, that no-storage claim conflicts with the supplied annotations (readOnlyHint=false, destructiveHint=true), so the agent receives contradictory signals about persistence and side effects.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two tight sentences with the operation front-loaded and the no-storage caveat trailing. The "Tuatara Topics:" prefix is dead weight but otherwise nothing is wasted.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no output schema, the description carries the burden of explaining returns and does so briefly (topics plus weights). It still leaves open whether num_words affects topic granularity and how the model's topic set is defined or changed, which matters for correct invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so both parameters are already documented, and the baseline is 3. The description adds only marginal meaning, clarifying that the document is scored against a specific "served model's topics" set and that results come back weighted, but says nothing about how num_words changes the output.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
"Place one document on the general topic set ... return the topics it belongs to with their weights" gives a concrete verb+resource and states the output shape, so an agent can tell this is a single-document topic-ranking call. It never names a sibling (e.g. find_topics, summarize_document, query_knowledge), so the differentiation is inferred rather than stated.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
There is no explicit when-to-use or when-not-to-use guidance and no named alternative, despite several plausible siblings (find_topics, summarize_document, query_knowledge). "one document" and "general topic set" hint that a custom topic set lives elsewhere, but the agent must guess that routing.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
find_topicstopic_of_documentSort up to 1,000 texts into your own labels with a calibrated confidence per answer. No API key.
Train a text classifier from your labelled examples in a minute; label texts with probabilities.
Search + read the web for $0.002/page; failed URLs free; per-URL errors never fail the call.
Turn documents into structured, AI-ready data by parsing, enriching, chunking, and embedding.
Add one secure layer between your agents and this server.
