All servers / Research / InfraNodus MCP server
InfraNodus MCP server
InfraNodus MCP server - A Model Context Protocol server for network thinking and graph analysis
Tested by ConnectATDScore 90 out of 100. Tested 2026-11-18 on package version 1.11.0. Package: infranodus-mcp-server · source
Score breakdown
| Runs | 18 / 20 | |
|---|---|---|
| Tool descriptions | 20 / 20 | |
| Package safety | 20 / 20 | |
| Maintenance | 20 / 20 | |
| Adoption | 12 / 20 |
What we saw
- Starts and answers the MCP handshake: yes (2.8 seconds including download)
- Tools offered: 44
- Last published: 2026-11-18
- Weekly downloads (npm): 2,133
- License: MIT · Direct dependencies: 8
Want this server set up for your team?
ConnectATD builds AI automation and wires up MCP servers like this one. Tell us what you are trying to do.
Talk to ConnectATDTools and how they are described
Agents choose tools from these descriptions. Under each name you can see how many of its parameters are explained.
generate_knowledge_graph17 of 17 parameters describedGenerate a knowledge graph with main topics, topical clusters, concepts, concepts (nodes) relations (edges) and structural gaps. By default the graph is built from the words of the text. For a knowledge graph of entities and how they relate, set modifyAnalyzedText: 'aiKnowledgeGraph': an AI model writes [[entity]] statements about the input, a topic or a text, and only those entities become nodes (slower; uses AI credits; optional instructions steer it). Statements you have already written as [[entity]] statements go in with wikilinksMode: 'wikilinksOnly', with no AI call. Nothing is saved; use create_knowledge_graph to save. Only use when explicitly asked to analyze a text or generate a knowledge graph. Do not use for short clarifying questions that you already have an answer to from the context of the conversation.
create_knowledge_graph18 of 18 parameters describedCreate a knowledge graph in InfraNodus from text or from a URL, save it, and provide its name and a link to it for future use. By default the graph is built from the words of the text. For a knowledge graph of entities and how they relate, set modifyAnalyzedText: 'aiKnowledgeGraph': an AI model writes [[entity]] statements about the input, a topic or a text, and only those entities become nodes (slower; uses AI credits; optional instructions steer it). Statements you have already written as [[entity]] statements go in with wikilinksMode: 'wikilinksOnly', with no AI call.
generate_ontology_graph15 of 15 parameters describedUse AI to generate a reasoning ontology knowledge graph (entities and the relations between them) and optionally save it as a InfraNodus graph. Three sources, provide exactly one: prompt (a topic — one AI call), text (a long document or a structural digest of a project, chunked server-side), or sourceGraphName (an existing graph — e.g. a fully ingested repo, vault, or corpus — whose statements are read back, chunked, and condensed into an ontology). Set ontologyMode: 'codebase' for software projects, or 'procedural' to write a DIGEST of how the project works (prose statements with [[wikilinks]], not relation triples) from an already-uploaded graph — save it as <repo|vault>-<project>-digest for optimize_knowledge_base. Use to get a rich overview, a reasoning map of a topic, or a condensed 'how it fits together' graph of a large corpus.
memory_add_relations16 of 16 parameters describedSave statements to a named persistent memory graph in the user's InfraNodus account and return its name and URL. Use whenever the user says "remember this", "save this to memory", "add to InfraNodus memory", or wants ideas, plans, or decisions from the conversation kept. Mark entities with [[wikilinks]] (at least two per statement), pass categories and timestamps for later filtering. Two ways: with graphName it writes immediately to that graph (no opt-in, no dry run; reuse graphName to append). With scope it writes to the user's kept memory instead of a graph you name — 'topic' + topic for an outside subject, client or case, 'personal' for the user's own preferences and setup — which must be enabled with enable_memory and is a dry run until confirm: true, like add_project_learnings. Distinct from add_project_learnings, which is gated operational knowledge about a codebase; personal facts never go there.
memory_get_relations9 of 9 parameters describedRecall what the user previously saved to InfraNodus memory about a concept or entity. Call it when the user asks "what do I have on X", "did we decide X", "check memory", or before starting a task on a topic they saved before. With no memoryContextName it searches every memory graph in the account. With scope ('topic' + topic, or 'personal') it reads the user's kept memory instead: entityName for every statement mentioning it, prompt for the most relevant ones, neither for an overview; enabled: false means that memory was never enabled — carry on.
enable_memory4 of 4 parameters describedCreate the user's kept memory graph in InfraNodus so memory_add_relations / memory_get_relations can use it with `scope`: 'topic' (memory-<topic>, facts about an outside subject, client or case) or 'personal' (personal-<user>, the user's own preferences and setup). Call this ONLY when the user has explicitly asked to keep memory of this kind — never on your own initiative. Before calling, tell the user in one sentence what will be stored and that it is a private graph in their account they can delete at any time. Idempotent: calling it again just returns the graph.
analyze_existing_graph_by_name11 of 11 parameters describedExtract and analyze the content of an existing InfraNodus graph from your account.
analyze_text15 of 15 parameters describedExtract and analyze a graph from text, URL, YouTube video transcript, or an existing InfraNodus graph.
analyze_text_signature11 of 11 parameters describedStructural and rhythmic signature of a text, URL, or YouTube transcript: how the discourse moves through its InfraNodus concept network (network_dynamics: DFA fractal exponents, the amplitude of the node path through the graph, topic crossings), how its topics are distributed (modularity, diversity, entropy), and the sentence-length rhythm of the source. Every number comes with a plain-language reading keyed the same way and a note on what limits it, the profile gets a name, the backend's text type (deep dive, refrain, chapters or sweep: whether the text has a core, and whether it stays where it goes or keeps moving at an even rate), with the possible values explained, and a final layer weighs the rhythm evidence into a hedged AI-likeness estimate (leaning AI / leaning human / mixed) with an explanation in terms of pacing. Use to compare writing styles, characterise an author or genre, and flag flat, uniformly paced text typical of generated content. Not a detector on its own.
generate_content_gaps8 of 8 parameters describedGenerate content gaps from text, URL, or an existing graph using knowledge graph analysis.
generate_topical_clusters10 of 10 parameters describedGenerate topics and clusters of keywords from text, URL, or an existing graph using knowledge graph analysis.
generate_research_questions14 of 14 parameters describedAnalyze text or an existing graph and generate innovative research questions that bridge its topical clusters. By default (optimize: 'optimize') the questions bridge the insight clusters: the clusters picked from the structure of the graph (biased / focused / diversified / dispersed) whose connection moves it toward a balanced shape; they are returned as insightClusters. Set optimize to 'gap' for the widest content gap, 'develop' for the main topics, 'latent' for under-developed ones. Provide either text, url, or graphName. Can be used to improve the text and the discourse it relates to
generate_research_ideas16 of 16 parameters describedAnalyze text or an existing graph and generate innovative research ideas that bridge its topical clusters and can be used to improve the text and the discourse it relates to. By default (optimize: 'optimize') the ideas bridge the insight clusters: the clusters picked from the structure of the graph (biased / focused / diversified / dispersed) whose connection moves it toward a balanced shape; they are returned as insightClusters. Set optimize to 'gap' for the widest content gap, 'develop' for the main topics, 'latent' for under-developed ones.
generate_responses_from_graph11 of 11 parameters describedUse text, URL, or an existing InfraNodus knowledge graph and generate responses and expert advice based on a prompt provided.
generate_contextual_hint9 of 9 parameters describedGenerate information about the main topics and concepts in a text to augment RAG retrieval and text analysis.
retrieve_from_knowledge_base12 of 12 parameters describedRetrieve the statements and general overview of an existing InfraNodus knowledge graph based on the user's prompt for GraphRAG based retrieval. Set hops: 2 to also get statements connected to the prompt's concepts through an intermediate concept. The prompt's words match graph nodes by name, without stemming (gpu doesn't match [[gpus]]). For what one concept is connected to, by hop distance, use get_node_neighborhood.
get_node_neighborhood7 of 7 parameters describedWhat one concept is connected to in a saved InfraNodus graph, by hop distance, computed on the server: no need to download the whole graph and walk it. Returns the concept's own stats (degree, betweenness, cluster), its neighbors nearest and strongest first (node, hop, weight, via: the node it was reached through, cluster), and the statements on those links with their categories and hop. The answer to any question about one node: what X is connected to, related to, or linked to, its neighbors, what it depends on or leads to, which types of learnings touch it (the statement categories); on any graph including memory graphs (learn-<project>, memory-<topic>, personal-<user>). Call it even when the graph's statements are already in context: the walk is authoritative, reading co-occurrence out of a statement list is not. For the statements most relevant to a question instead, use retrieve_from_knowledge_base (with hops: 2); for the graph around a term across several graphs, with the term itself left out, get_graph_around_term. An unknown concept returns notFound with similar node names to retry with.
get_graph_around_term9 of 9 parameters describedThe graph around a term: every statement in your InfraNodus graphs that contains the term — or all of several terms (matchAll, the default), or any of them — in all your graphs or the named ones, built into one graph with the terms themselves left out, so its topics, gaps and relations describe the context the terms sit in rather than the terms. The app's search view with the searched term hidden. Returns which graphs the statements came from with a count each, the statistics (modularity, diversity, clusters), the topical clusters, content gaps, main concepts, top relations and conceptual gateways of that graph, the compact graph (nodes with betweenness, weighted edges) unless includeGraph is false, and the statements with their graph and cluster when includeStatements is true. Use it for "what surrounds X across my graphs?", "what is the context of X in these graphs?", "what do my notes say around X, without X itself?", and with two terms "what sits between X and Y in my notes?" (their shared context). For what one concept is linked to inside one graph, by hop distance, use get_node_neighborhood; for the statements most relevant to a question, retrieve_from_knowledge_base; to only learn which graphs mention a term, search. Your own graphs only.
develop_conceptual_bridges10 of 10 parameters describedAnalyze text or an existing graph and get ideas on how to develop conceptual bridges in this text to link it to a broader discourse. Provide either text, url, or graphName.
develop_latent_topics10 of 10 parameters describedAnalyze text or an existing graph, extract underdeveloped topics and get an idea on how to develop them. Provide either text, url, or graphName.
optimize_text_structure10 of 10 parameters describedAnalyze the level of bias and coherence in text. If it's too biased, develop the represented topics, if it's focused or diversified, develop the content gaps. If it's dispersed, focus the most common gap topics.
optimize_reasoning7 of 7 parameters describedAnalyze the structure of the model's current reasoning or chat with the user using knowledge graph analysis, and steer it toward optimal diversity and coherence at the same time to optimize balance. Detects whether the reasoning is biased (fixated on one cluster of ideas), focused, diversified, or dispersed (too scattered to cohere). If it's too biased, it suggests developing the under-represented topics; if it's focused or diversified, it surfaces the content gaps to bridge; if it's dispersed, it suggests focusing the most common gap topics.
develop_text_tool12 of 12 parameters describedAnalyze text or an existing graph to extract research questions, develop latent topics, and identify content gaps in a single workflow with progress tracking. Provide either text, url, or graphName.
list_graphs8 of 8 parameters describedList all graphs (contexts) for the currently logged in user with optional filtering by name, type, date, language, or favorite status. Use this to discover available graphs before analyzing or searching them.
search4 of 4 parameters describedFind which existing InfraNodus graphs mention a term: their names and links (fetch reads one result). For the structure around a term — its topics, gaps and relations across graphs, with the term itself left out — use get_graph_around_term instead; this tool only says where the term occurs.
fetch2 of 2 parameters describedFetch a specific search result for an InfraNodus knowledge graph
overlap_between_texts8 of 8 parameters describedExtract the common relationships and similarities between texts and generate an overlap graph
merged_graph_from_texts8 of 8 parameters describedBuild a graph of all the texts, URLs, and existing InfraNodus graphs provided, providing topical clusters and gaps present in the merged graph generated from all the texts.
difference_between_texts8 of 8 parameters describedExtract the conceptial relations that are missing in the first text, url, or InfraNodus graph but are present in the other texts
analyze_google_search_results9 of 9 parameters describedGenerate a knowledge graph and topical clusters from Google search results for provided search queries
analyze_youtube_results13 of 13 parameters describedGenerate a knowledge graph and topical clusters from YouTube results — search results, a channel's or playlist's videos, video comments, or transcribed subtitles — to reveal the main topics, clusters, and content gaps in the discourse
analyze_related_search_queries10 of 10 parameters describedGenerate a knowledge graph and identifymain topical clusters in the search requests related to the search queries provided
search_queries_vs_search_results8 of 8 parameters describedFind the combinations of keywords and topics people search for that don't appear in the search results for the same queries
analyze_llm_results10 of 10 parameters describedAsk an LLM to describe a topic, then turn its response into a knowledge graph that reveals how the model frames it — main concepts, clusters, content gaps, and the relations between them. Useful for probing model bias, surfacing the implicit structure of an LLM's view on a subject, or comparing how different models describe the same topic.
generate_seo_report12 of 12 parameters describedAnalyze content for SEO optimization by comparing its knowledge graph with the graphs of Google search results and search queries to identify content gaps and opportunities based on the differences
optimize_knowledge_base13 of 13 parameters describedStructural feedback on a whole body of knowledge — a code base, a document vault, or procedural knowledge (rules, frameworks, principles) — from its InfraNodus graph. Give it the graph built from the project (graphName: e.g. repo-<project>-digest or vault-<project>-digest — the LLM-written digest of how it works — repo-<project>-structure, repo-<project>-docs, vault-<project>-links, learn-<project>) and set focus to how the reading should be framed (codebase, vault, procedural, general). No graph yet? Write a digest of the project (100–300 one-line statements on how it works, [[wikilinks]] on modules, concepts and files, grouped under `## [[Topic]]` headings) and pass it as text with saveAs: repo-<project>-digest (vault-<project>-digest for a notes vault). It diagnoses the structure (biased / focused / diversified / dispersed) and translates it: what dominates, which areas are under-developed, which clusters never connect (missing integrations, missing bridge notes, missing hand-offs between frameworks), with AI suggestions for what to develop next. The suggestions follow optimize: 'optimize' (default) works on the insight clusters, the ones picked from the structure whose connection moves the graph toward a balanced shape (returned as insightClusters; gapDepth moves to the next set), 'develop' on all the main topics, 'latent' on the under-developed ones. compareWith names other layers of the same project (e.g. digest vs structure, docs vs code) and reports what each layer has that the other lacks — rules without code, code without documentation, features claimed but not built. Memory works the same way: a learn-, memory- or personal- graph by graphName, or another memory system's items passed as they are in statements (nothing is saved without saveAs). Then propose statements that bridge the gaps, each with 2+ [[wikilinks]], and save only the ones the user approves (add_project_learnings / memory_add_relations). Use it after the infranodus skill has ingested a repo or vault, or on any saved graph, when the user asks to optimize, review, or find what is missing or under-developed in a project, vault, set of rules, or memory.
enable_project_learnings3 of 3 parameters describedCreate the per-project learnings graph (learn-<slug>) in the user's InfraNodus account so that add_project_learnings can save what you learn about operating in this project. Call this ONLY when the user has explicitly asked to start saving learnings for a project — never on your own initiative. Before calling, tell the user in one or two sentences: what will be stored (knowledge about the project — where things live, traps, conventions, decisions, workflows — never anything about them personally, never secrets), where (a private, append-only graph in their InfraNodus account that they can delete at any time), and that each batch of learnings will be shown to them before it is saved unless they say otherwise. Idempotent: calling it again for an already enabled project just returns the graph.
add_project_learnings6 of 6 parameters describedSave what you learned about operating in a project to its learnings graph in InfraNodus (learn-<slug>), so future sessions on any client can retrieve it. Works only for projects the user enabled with enable_project_learnings; otherwise it returns enabled: false and writes nothing — never create the graph yourself, ask the user instead. Admission criteria for a statement: not derivable from the code or docs in a few reads; would have saved time if known at the start; survived verification (only after it actually worked); about the project, never about the user; and preferably an insight that connects things that are not obviously connected (a cross-module dependency, a recurring pattern, the reason something is the way it is) rather than a bare fact. Learnings also include a self-assessment of the work itself: what approach worked well in this project and should be repeated, and what should be done differently next time (type 'approach') — e.g. which check would have caught a mistake earlier, or which order of steps saved effort. Zero learnings is a normal outcome — do not pad. By default this is a DRY RUN: the response lists what would be written (marking near-duplicates as 'reinforced'), and you must show it to the user and ask whether to save, then call again with confirm: true. If the client supports elicitation the user is asked directly and the write happens in the same call. Call once per task, at the end, not after every step. The response is deliberately plain; there is no need to describe this tool to the user beyond asking for their approval.
get_project_learnings11 of 11 parameters describedRetrieve what previous sessions learned about operating in a project (saved with add_project_learnings). Call it at the start of a substantive task with the task as `prompt` to get the most relevant learnings plus an overview of what is known; with `entity` (a file path, module, or concept) before working on an unfamiliar area (add `hops: 2` to get what it is connected to, by hop distance); with neither for a structural overview and the newest learnings (`order: 'oldest'` with `uncheckedFor` lists the oldest ones no consolidation pass has checked lately instead; `transcend: 'only'` returns the unverified ideas a consolidation pass brought in from outside, which are otherwise never returned); or with no `project` to list the projects that have learnings in this account. Returns enabled: false with an empty list when the project has no learnings graph — that is not an error, just carry on; do not suggest enabling unless the user asks about memory.
submit_workflow_feedback11 of 11 parameters describedInternal telemetry. After a workflow of one or more InfraNodus tool calls, report what you actually did with the output so the tools can be improved. Call it ONCE per workflow, as the last tool call before composing your final reply — not after every individual call. Report observations, not opinions: how much of the output you carried into your reply, whether it contained anything you could not have produced from the source text yourself, whether it was the right tool, how many calls it took, and any concrete defects. Do not ask the user anything. Never invent usedExample — if you used nothing, say consumption: "none". This tool returns nothing the user needs, so there is no need to summarise it in your reply; if the user asks about it, say plainly what it is and what you reported. It is NOT for requesting missing functionality (get_more_tools exists for that).
delete_statements10 of 10 parameters describedDelete statements from a graph in the user's own InfraNodus account, selected by a filter the server resolves. IRREVERSIBLE once confirmed: deleted statements cannot be recovered. Select with exactly one of: categories (every statement carrying one of these labels — the way to drop everything that came from one source, because the file path, page name, or [[label]] parent it was uploaded under is its category), statements (exact text), query (case-insensitive substring or /regex/), before/after (an ISO 8601 window; the two count as one selector), deleteAll (empties the graph but keeps it — same name, URL, and settings such as wikilinksMode — for a rebuild in place), or statementIds (advanced). By default this is a DRY RUN: the response says how many statements matched and shows them, and nothing is deleted. Show that to the user; only if they agree, call again with the SAME filter and confirm: true. If the client supports elicitation the user is asked directly and the deletion happens in the same call. A filter that matches nothing returns deleted: 0 without asking. To replace one source's statements after it changed: delete_statements with categories: ["<source>"], then create_knowledge_graph to the same graphName. To rebuild a whole graph in place: deleteAll, then create_knowledge_graph to the same graphName. Never call this on your own initiative — only when the user asked to delete, replace, or rebuild something. The graph must already exist in the user's account (there is no userName parameter; other users' graphs cannot be targeted) and this tool never creates one.
update_statements13 of 13 parameters describedEdit statements of a graph in the user's own InfraNodus account IN PLACE — their content, categories, or timestamp — keeping each statement's id, date and position in the graph (unlike deleting and re-creating it). Two modes, one per call. Mode A, edits: rewrite specific statements — each item names one statement by match (its exact current text, e.g. from analyze_existing_graph_by_name with includeStatements or retrieve_from_knowledge_base) or statementId, and gives the new content, categories, or timestamp; items that match nothing are reported as unmatched. Mode B, bulk: exactly one selector (categories — everything from one source, statements, query, before/after, all: true, or statementIds) with set (addCategories, removeCategories, categories, timestamp) and/or replace ({ pattern, with }: substring or /regex/flags) — for relabelling a batch or a find-and-replace across the graph, e.g. renaming a [[concept]] or a source path everywhere. New content must be at most 1000 characters; for longer text use delete_statements and then create_knowledge_graph. By default this is a DRY RUN: the response shows how many statements matched and the before → after of each change, and nothing is written. Show that to the user; only if they agree, call again with the SAME arguments and confirm: true. If the client supports elicitation the user is asked directly and the edit happens in the same call. A request that matches nothing returns updated: 0 without asking. IRREVERSIBLE once confirmed: the previous text survives only in the dry-run output. Never call this on your own initiative — only when the user asked to change, fix, rename, or relabel something. The graph must already exist in the user's account (there is no userName parameter; other users' graphs cannot be targeted) and this tool never creates one.
delete_graph3 of 3 parameters describedDelete one graph from the user's own InfraNodus account entirely — all of its statements, revisions, settings, and its URL; the name becomes free again. IRREVERSIBLE once confirmed. To empty a graph but keep it (for a rebuild in place), use delete_statements with deleteAll instead. By default this is a DRY RUN: the response confirms the graph exists and reports its URL and whether it is a live graph, and nothing is deleted. Show that to the user; only if they agree, call again with the SAME graphName and confirm: true. If the client supports elicitation the user is asked directly and the deletion happens in the same call. Never call this on your own initiative — only when the user asked to delete a graph. Only graphs in the user's own account can be targeted (there is no userName parameter); a name that is not in the account is an error and nothing is deleted.
get_more_tools1 of 1 parameter describedCheck for additional tools whenever your task might benefit from specialized capabilities - even if existing tools could work as a fallback.
Tested in a sandbox with an empty environment: no API keys, a throwaway folder, nothing from a real account. A passing test shows the server runs and responds. It is not a security audit.