Skip to main content
Glama
LiLara-AI

ShadowGraph

Official

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
SHADOWGRAPH_FILENoAbsolute path to the ShadowGraph data file. Use when one store must be shared across working directories. Defaults to `.shadowgraph/data.json`..shadowgraph/data.json
SHADOWGRAPH_STORAGENoStorage backend: 'json' (default) or 'sqlite' (requires Node 22.5+). Set to 'sqlite' to use the WAL-backed relational adapter.json
SHADOWGRAPH_MCP_COMPACTNoSet to '1' to enable compact mode, which advertises 12 workflow tools instead of 27. Default is '0' (full mode).0
SHADOWGRAPH_ALLOW_REMOTE_EMBEDDINGSNoSet to '1' to allow remote embedding endpoints, sending memory and query text off the machine. Off by default; localhost OpenAI-compatible servers work without this.0

Instructions

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

This server publishes no instructions, or was last inspected before Glama recorded them.

Capabilities

Features and capabilities supported by this server

Protocol revision2025-11-25

CapabilityDetails
tools
{
  "listChanged": false
}
prompts
{
  "listChanged": false
}
resources
{
  "subscribe": false,
  "listChanged": false
}

Tools

Functions exposed to the LLM to take actions

NameDescription
shadowgraph_record_decisionA

Record one decision: chosen option, assumptions, evidence, and rejected alternatives with reopenWhen rules. Use shadowgraph_record_attempt for something tried, shadowgraph_record_fact for an observation, shadowgraph_remember for a note. Appends a decision and a journal entry; without idempotencyKey each call adds another and commits a revision.

shadowgraph_record_attemptA

Record one attempt and how it turned out, so the same approach is not blindly retried. Use shadowgraph_record_decision for the choice itself, shadowgraph_record_outcome for how one played out. Appends an attempt and a journal entry; without idempotencyKey a retry records a second one.

shadowgraph_reviewA

List decisions whose rejected alternatives are due again, from reopenWhen rules over stored facts. shadowgraph_review_signals reads persisted ones, shadowgraph_ack_review closes one, shadowgraph_maintain ages first. Persists one signal per newly due decision, deduped by decision and reason; a repeat commits a revision. Returns a bare JSON array.

shadowgraph_searchA

Find decisions, attempts, and facts whose declared content fields contain every query term. shadowgraph_retrieve adds one-hop neighbours, shadowgraph_recall fuses ranks over scoped memory, shadowgraph_context builds a working set, shadowgraph_traverse walks from an id. Reads only. Schema keys never match, and a filter is never a content match.

shadowgraph_contextA

Build one project's working set before a consequential task: decisions, stale assumptions, failed attempts, open reviews. shadowgraph_search or shadowgraph_retrieve look one thing up, shadowgraph_recall reads scoped memory, shadowgraph_review only evaluates. Not a read: it evaluates reopen rules, can persist signals, and commits a revision.

shadowgraph_rememberA

Add or reconcile one scoped memory, or apply an ADD/UPDATE/DELETE/NOOP plan, by identity tuple. Use shadowgraph_record_decision for a choice, shadowgraph_record_fact for an observation, shadowgraph_recall to read memory back. Identical content is a NOOP, new content supersedes and keeps history, DELETE invalidates; every call commits a revision.

shadowgraph_recallA

Recall scoped memory and records by fusing lexical, vector, graph-distance, and temporal ranks. shadowgraph_search is plain content matching, shadowgraph_retrieve adds graph neighbours, shadowgraph_remember writes memory. Reads only. Every response declares which signals were available, so absent ones are never renamed.

shadowgraph_record_factA

Record one observed fact as a project and key, with a claimed provenance class and optional validity window. Use for the fact keys reopenWhen rules name; shadowgraph_remember stores durable memory that is not an observation. Supersedes the previous active fact, keeping it as history. No input can make a fact verified. Each call commits a revision.

shadowgraph_record_outcomeA

Record how a decision turned out and move its confidence by an evidence-weighted amount. shadowgraph_confidence_evidence records evidence short of an outcome, shadowgraph_update_status changes the lifecycle state. One outcome contribution per decision, so re-recording replaces rather than stacking, restamps it, and commits a revision.

shadowgraph_confidence_evidenceA

Apply one keyed piece of supporting or contradicting evidence to a decision's confidence. Use shadowgraph_record_outcome once the decision has played out, shadowgraph_record_fact for an observation that can reopen it. Reusing key cannot double-count, but restamps the decision and commits a revision; a new observation needs a new key.

shadowgraph_update_statusA

Move one decision to another lifecycle state. shadowgraph_supersede replaces a decision, shadowgraph_maintain alone produces stale, shadowgraph_record_outcome records the result. stale and superseded are system-owned and rejected. Setting the current state writes nothing but still commits a revision.

shadowgraph_linkA

Create one named, directed relationship between two entities that already exist. Read relationships back with shadowgraph_traverse; shadowgraph_supersede records replacement and its own relation. Every call mints a new relation id, so a repeat duplicates the relationship. There is no unlink tool.

shadowgraph_traverseA

Walk relationships outward from one entity id and return what is reached. Find ids first with shadowgraph_search or shadowgraph_recall; shadowgraph_retrieve searches content and adds neighbours. Reads only. Memory outside the requested project and scope stays hidden.

shadowgraph_supersedeA

Mark one decision superseded by a replacement in the same project. shadowgraph_update_status handles ordinary lifecycle moves, shadowgraph_link any other relationship. Nothing is deleted: the previous decision stays searchable. A repeat returns the same result and commits a revision.

shadowgraph_redactA

Return a redacted copy of the store export, with secret-looking keys and values replaced. shadowgraph_backup writes an unredacted snapshot to disk, shadowgraph_purge actually removes data. Reads only, writes no file, and redacts journal payloads too, so a secret cannot survive in the audit trail.

shadowgraph_purgeA

Delete one project's decisions, attempts, memories, facts, relationships, events, and retry keys. Run shadowgraph_purge_preview first; shadowgraph_redact shares without deleting, shadowgraph_backup keeps a copy. Irreversible without a backup. logical keeps an auditable journal skeleton; hard also drops entries, leaving a declared gap.

shadowgraph_maintainA

Run time-based maintenance: make due decisions stale, expire due facts, then evaluate reopen rules. shadowgraph_review only evaluates reopen rules, shadowgraph_validate only reports, shadowgraph_update_status cannot set stale. Writes, and depends on the clock: a repeat at the same instant finds nothing to do but still commits a revision.

shadowgraph_retrieveA

Retrieve records matching a content query together with their one-hop graph neighbours. shadowgraph_search returns matches only, shadowgraph_recall ranks scoped memory, shadowgraph_traverse walks from a known id, shadowgraph_context builds a project working set. Reads only. A neighbour can appear with no content match of its own.

shadowgraph_validateA

Report graph integrity diagnostics by severity, without modifying storage. shadowgraph_repair_plan shows what a fix would involve, shadowgraph_rebuild tests whether the journal reproduces the data. Reads only. Legacy data is named rather than guess-repaired, and newer-schema data is reported, not downgraded.

shadowgraph_journalA

Read the append-oriented journal of complete post-operation snapshots, in sequence order. shadowgraph_rebuild replays it into a projection, shadowgraph_validate diagnoses live data, shadowgraph_search finds records. Reads only. Entries are immutable audit evidence and are never rewritten in place.

shadowgraph_rebuildA

Replay this store's own journal into a projection and report whether the fold was complete. shadowgraph_journal reads the entries themselves, shadowgraph_validate diagnoses the live graph. Reads only: the journal is untouched and the live graph is not replaced by the result.

shadowgraph_review_signalsA

List the review signals already persisted, optionally narrowed to a project or to open or acknowledged ones. shadowgraph_review re-evaluates reopen rules and can create signals, shadowgraph_ack_review closes one. Reads only. Acknowledged signals are retained rather than deleted. Returns a bare JSON array, not paginated.

shadowgraph_purge_previewA

Count what a purge of one project would remove, without changing anything. Run this before shadowgraph_purge; it takes the same project name and reports the same counts. Reads only: no storage, journal, or signal change, and no file is written.

shadowgraph_ack_reviewA

Acknowledge one persisted review signal by id, closing it as open work. Ids come from shadowgraph_review_signals or shadowgraph_review; use shadowgraph_update_status or shadowgraph_supersede to act on the decision. Overwrites status and acknowledgedAt in place with no journal entry, so a rebuild cannot reconstruct it; a repeat restamps it.

shadowgraph_repair_planA

Return the non-destructive repair plan implied by the current diagnostics. Use after shadowgraph_validate; there is no apply tool, so the caller carries out every action with ordinary tools. Never mutates: the result always carries apply:false, and anything ambiguous is routed to manual_review.

shadowgraph_backupA

Write a consistent snapshot of the whole store to a filesystem path on the server. Take one before shadowgraph_purge or shadowgraph_restore; shadowgraph_redact shares data without writing a file. Overwrites any existing file at destination without warning, creates parent directories, and commits a revision.

shadowgraph_restoreA

Replace the entire live store with the contents of a JSON or SQLite backup. Recovery only, never a merge: shadowgraph_purge removes one project, shadowgraph_backup makes the snapshot this reads. Destructive: every record and journal entry is replaced and a strictly greater revision installed. A failure is rolled back.

Prompts

Interactive templates invoked by user choice

NameDescription
shadowgraph_consequential_taskUse ShadowGraph before, during, and after consequential work.

Resources

Contextual data attached and managed by the client

NameDescription
ShadowGraph contextCurrent project context and open review signals.

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/LiLara-AI/shadowgraph'

If you have feedback or need assistance with the MCP directory API, please join our Discord server