graphrag-ts is a TypeScript GraphRAG engine on PostgreSQL/pgvector, served as a Streamable HTTP MCP endpoint. Public read tier needs no credentials (20 req/min per IP): integrity_paper_status, retraction_propagation, integrity_evidence_audit. A bearer key unlocks graph query, global map-reduce, index jobs and memory writes. Free browser demo: https://huggingface.co/spaces/ashesborn6/retraction-status
Scored across 8 tools
Most tools have clearly distinct purposes, and the descriptions explicitly separate close pairs like query_graph vs query_graph_global and submit_index_job vs remember_text. The integrity family is also differentiated, though integrity_paper_status vs integrity_evidence_audit could be confused at first glance by an agent scanning names alone.
The naming mixes verb-first tools such as get_index_job, submit_index_job, query_graph, and remember_text with noun-first or domain-prefixed tools like integrity_paper_status, integrity_evidence_audit, and retraction_propagation. The groupings are readable and semantically meaningful, but the overall convention is not uniform.
Eight tools is a well-scoped set for the combined indexing/query and integrity-analysis surface. Each tool appears to earn its place, with no obvious redundant operations.
The core lifecycle is covered: ingestion via submit_index_job or remember_text, polling via get_index_job, and querying via local or global retrieval. Integrity workflows move from paper status to evidence audit to propagation, though there are minor gaps such as no workspace/document listing, job cancellation, deletion, or snapshot inspection.
Get the current state of an indexing job: lifecycle status, build phase, timestamps, the published snapshot version when one exists, and a safe error summary when it failed. Use it to poll after submit_index_job or remember_text, and to confirm that a new snapshot is live before querying. Status values: queued, in_progress, published, failed. Returns: job_id, workspace_id, status, phase, created_at, started_at, finished_at, error {code, summary} or null, published_version {id, workspace_id} or null. Errors: JOB_NOT_FOUND when no job with that id exists in the workspace.
Inputs: job_id (required), workspace_id.
Audit the evidence behind a paper's retraction picture: every source record with its tier and URI, the dates involved, conflicts between sources (an official registry and an aggregator disagreeing, or a citation acknowledging a retraction the graph does not propagate) and the coverage gaps that bound the conclusion. Use it when you must justify or challenge a verdict; for a one-line status use integrity_paper_status. Returns: evidence records with source tiers and URIs, detected conflict flags with codes and details, and the coverage that limits how strong the conclusion can be. Paid: $0.10 per call (x402 on Base, POST /api/v1/judge); a keyed MCP session also unlocks it.
Inputs: doi, paper_id.
Look up one paper by DOI or OpenAlex id and return its identity (title, year, venue) plus every retraction record the graph holds for it, each with its source and source tier: A is an official registry, B an authoritative index, C an aggregator. Use it to check whether a paper is retracted and which sources assert it; use retraction_propagation to judge whether that retraction reaches a specific citing paper. Free in the public tier. Returns: paper identity and statuses[] with source, source_tier, is_retracted, retraction_date and source_uri; abstention_reason when the paper is not in the snapshot.
Inputs: doi, paper_id.
Answer a question over one workspace's knowledge graph with local hybrid retrieval: vector similarity, keyword match and graph neighbourhoods are combined, the strongest passages and their entities are selected, and the answer is composed from them with source evidence. Use it for questions a handful of documents can answer; use query_graph_global for corpus-wide synthesis. It reads the latest published snapshot only, so results are stable until a new build publishes. Returns: an answer plus the evidence snippets behind it, each with the source document and chunk it came from.
Inputs: query (required), max_results, workspace_id, community_level.
Answer a corpus-wide question by Map-Reduce over community summaries: every selected community is summarised against the query, then the partial answers are reduced into one response that lists the communities it drew on. Use it for themes, trends and overviews across a whole corpus; for pinpoint lookups or single-document questions use query_graph instead. Returns: the reduced answer with the community summaries and the source evidence behind them. Raises QUERY_CAPACITY when the server is at its active-query limit — retry shortly.
Inputs: query (required), workspace_id, community_level.
Queue caller-supplied text as a document in a workspace's knowledge graph. Use this when the text is not already on the server's filesystem: it indexes exactly the text you pass and never reads chat history, files or URLs. The text becomes citable source material for later query_graph answers. Indexing is asynchronous — follow up with get_index_job, and the text is not searchable until the job publishes a new snapshot. Returns: job_id, workspace_id, status ("queued"), queue_position.
Inputs: text (required), source, workspace_id.
Judge whether a retraction propagates to a paper through its citation graph: retraction evidence is weighted by source tier, the citing context is checked for an acknowledgement, and the tool returns CONTESTED, SUPPORTED or ABSTAINED with the evidence path behind the verdict. Use it after integrity_paper_status when the question is "does this retraction reach this paper". It abstains instead of guessing when reference coverage is thin. Returns: verdict, confidence, the evidence path that produced it, and abstention details when the verdict is ABSTAINED. Paid: $0.02 per call (x402 on Base, POST /api/v1/judge); a keyed MCP session also unlocks it.
Inputs: doi, paper_id.
Queue Markdown files from the configured read-only input directory for indexing into a workspace's knowledge graph. Returns immediately with a job id; the build runs in the background, so poll get_index_job and only call query_graph after a new snapshot is published. Files are snapshotted when the job is accepted; later edits on disk are not picked up by this job. Shared-filesystem ingestion is disabled on hosted endpoints — use remember_text there. Returns: job_id, workspace_id, status ("queued"), queue_position.
Inputs: path (required), source, workspace_id.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Shared memory for AI agents, as a graph in your own Postgres. Writes never call an LLM.
Provenance-carrying memory for AI agents: corrections supersede, never overwrite. Free, API + MCP.
Cloud or self-hosted knowledge for AI agents: hybrid search, reranking, GraphRAG, scoped MCP tools.
Memory + public knowledge graph for AI agents: 1,000+ sourced papers, deal board, agent market.
Add one secure layer between your agents and this server.
