dev.agentic-atlas/atlas
name:dev.agentic-atlas/atlas
Agent-design pattern guidance via 8 hosted read-only tools; Streamable HTTP, no auth, one Release.
- transport:
- remote
- credential class:
- self-provisionable
Owner verification
Not yet verified. Verifying proves you control this server and is free, permanently — it never changes a published score.
Start verification →Tools
- atlas_cardsshallow
Read 1-4 distinct canonical ids as one ordered Card per id. Use atlas_cards for a subject's gist or a comparison after you hold canonical ids; output order follows 'ids'. Each Card carries identity, its Hook (the one-line why-it-matters summary), and sealed decision-bearing claims with compact source addresses. The batch is atomic: every id must be distinct and readable in the same Release or 'batch_not_atomic' refuses the whole request. Use atlas_read instead for exact Section wording or a complete multi-Section Node; Cards omit Sections, links, and expanded provenance. Those omissions appear in 'scope.omitted_fact_classes'; the surface recovery map routes them to atlas_read, atlas_links, or atlas_provenance at the Card's own id or cited addresses.
- atlas_decisionsshallow
List or search the admitted Decision summaries in manifest order. Each summary carries the Decision's id, title, and Hook, always in Export-manifest order. 'query' matches the canonical id, title, or full body; pass an exact id for its single summary, or omit 'query' to list every admitted Decision. An empty result is a successful domain miss. Use atlas_read at 'decision:<id>' only when the complete record is needed.
- atlas_defineshallow
Define one admitted glossary term, with match, ambiguity, or miss. The payload's 'match' field states the outcome. 'exact': the term — or a case-insensitive substring naming exactly one term — resolved, and 'term' plus 'definition' are carried. 'ambiguous': several terms contain the query; their names come back in 'terms', and calling atlas_define again with one of them resolves it. 'none': no admitted term matches — a successful miss, not an error. The vocabulary is the active Release's own glossary.
- atlas_linksshallow
Page one subject's relationship occurrences, inbound and outbound. Use atlas_links to inspect a known Node's graph without loading its Section prose; use atlas_read for the full Node or exact wording. Occurrences are outbound first in Section-and-link order, then inbound, and each states direction, reference class, source Section, ordinal, and the disclosed far-end identity and Hook. 'bound' sets the page size; continue 'next_cursor' with the same subject, bound, and expected revision. The page is never direction-filtered: read each occurrence's 'direction' instead.
- atlas_navigateshallow
Read the canonical tree or publisher-curated tour. 'tree' (the default) is the canonical Catalog traversal: every published node with its id, title, position, and depth, in the publisher's manifest order, never re-ranked. 'tour' is the publisher-curated first learning path: an ordered walk with an intro and per-step narration naming each stop's node id; a Release that seals no tour answers 'path': null as a successful miss, not an error. Use 'tree' for the corpus's whole shape and 'tour' for a guided first reading; use atlas_orient instead for relevance-ranked discovery and atlas_read when you need a Node's content.
- atlas_orientshallow
Discover canonical identity from an ordinary-language query. The entry point for a caller that does not yet hold a canonical id — a caller that does should call atlas_cards or atlas_read directly. Query terms match canonical id, title, Hook, and full published Section text, in that rank order, with manifest order breaking ties. Section text is a fallback only when id, title, and Hook produce no candidates, and a body-only hit requires all query terms. Roadmap disclosures match only their public id, title, and Hook. Each candidate carries id, status, Hook, and disclosure class. Results are paged: 'scope' declares total/returned/truncated and 'next_cursor' while more follow. A query matching nothing is a successful empty payload, not an error.
- atlas_provenanceshallow
Expand compact source addresses into their audit facts. The separately charged audit payload: for each cited span address it returns the span's dependency digest and every attestation relying on it — distillation id, publisher, and attestation date — plus the expanded identity of the Release itself (catalog revision, corpus commit, candidate digest, export manifest digest). Call it only when auditing how a claim is sourced; no ordinary payload carries these facts. The batch is atomic: a malformed, unadmitted, or uncited address refuses the whole request with 'batch_not_atomic' naming the rejected addresses.
- atlas_readshallow
Read one Node id, node#section, or decision:<id> address. Three address grammars, each selecting its payload class: a bare node id (for example 'ledger-offload') returns the complete Node — identity, hierarchy, Hook, every Section in order, and the full relationship set; 'node-id#section-slug' returns exactly that one Section's heading and content; 'decision:<decision-id>' returns one complete Decision record. Reach for a Section when the exact wording of one part is needed and for the whole Node only for genuinely multi-Section work — a Card from atlas_cards is usually sufficient and cheaper. An address naming nothing admitted is refused with 'invalid_argument'.
Embed this server’s score
Tool count and median score across every tool in this server’s corpus — honest in a way a single cherry-picked tool’s badge wouldn’t be.
[](https://vouch.tools/servers/1864c119-c7ac-4f44-a65e-b23ed768903f)