Skip to main content
Glama

recall

Read-onlyIdempotent

Searches stored memories by keywords to locate a specific fact or decision, returning ranked matches with optional full content.

Instructions

SEARCH memory by keywords. Finds a SPECIFIC piece of knowledge, e.g. 'API convention' or 'license decision'. Returns ranked results with scores. Default view is compact: per item key, tags, score and the first ~300 chars of content (a trailing … marks a cut; timestamps, scope and bookkeeping fields are omitted). get_memory(key) returns the whole text of one entry; full=true returns every item uncut with all stored fields (created_at included). Below the relevance threshold, returns an empty list with an explanation instead of weak matches. include_archived=true also returns archived memories. Use it to find one fact; recall_with_graph also follows links to related memories, get_project_knowledge pages through one project's permanent knowledge.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "query": "<query>"
    },
    "name": "recall"
  }
}

Input Schema

NameRequiredDescriptionDefault
fullNoReturn every item in full, with all stored fields and uncut content (default false: compact view, content cut at ~300 chars and marked with a trailing `…`, bookkeeping fields such as agent_id, workspace_id, simhash and decay scores are left out). Prefer get_memory(key) to read one cut entry in full; full=true re-fetches the whole page uncut.
tagsNoAND-filter: memory must contain ALL listed tags.
limitNoMax results (1–50). Omitted compact requests use 5; full=true or an explicit profile keeps the previous default 10. Existing profile presets still apply when limit is omitted (multi-session uses 20, including automatic classification). Explicit limit wins. This is a hard bound: the response never contains more than limit items. When knowledge-graph boost is enabled, a share of the page (limit/4, at least 1 when limit>=2) may be filled with graph-expanded neighbours, marked graph_boost=true and provenance=via:graph — they take the tail slots instead of being added on top. Rows that fail scope/tags are dropped, never returned unmarked, whether they arrived by retrieval, by pinning, or by graph expansion.
queryYesFull-text search query.
scopeNoFilter by scope: workspace, project, agent, or all (default).
sinceNoReturn memories created at or after this RFC3339 timestamp.
untilNoReturn memories created at or before this RFC3339 timestamp.
offsetNoPagination offset (default 0).
order_byNoSort order: created_at:desc (default), created_at:asc, relevance:desc, decayed_relevance:desc.
tags_anyNoOR-filter: memory must contain AT LEAST ONE of these tags.
created_byNoFilter by agent ID (UUID).
project_idNoFilter to a specific project.
relevance_minNoMinimum relevance score (0-1).
min_importanceNoMinimum importance_score threshold (0-1, default 0.3 — matches the lowest score the server assigns, kind:session-checkpoint, so prior-session hand-offs are returned without an override). Raise it to exclude low-value entries; pass 0 to disable filtering entirely.
recall_profileNoOverride automatic query classification. Existing preset values are preserved; explicit limit takes precedence.
include_expiredNoAlso return memories past their expires_at (default false). An expired entry is hidden from recall and get_memory the moment it expires and stays readable with this flag until the background cleanup (every 6 h) physically deletes it 7 days after expiry. To take an entry out of recall so it can be brought back, use archive_memory instead of a short expires_at.
include_archivedNoAlso return archived memories (retired by their owner or by review but kept for history). Default false: only live memories.
apply_recency_decayNoSort by relevance * 0.95^days_since_created.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed1 schema field changedv0.1.7
    • changedInput schema / properties / include_expired / description
      Previous value: -"Include expired memories (default false)."New value: +"Also return memories past their expires_at (default false). An expired entry is hidden from recall and get_memory the moment it expires and stays readable with this flag until the background cleanup (every 6 h) physically deletes it 7 days after expiry. To take an entry out of recall so it can be brought back, use archive_memory instead of a short expires_at."
  2. Changed1 schema field changedv0.1.6
    • changedInput schema / properties / include_archived / description
      Previous value: -"Include archived memories in results (default false)."New value: +"Also return archived memories (retired by their owner or by review but kept for history). Default false: only live memories."
  3. First observedv0.1.1

TDQS

A4.3/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already cover safety (readOnly, idempotent, non-destructive, closed-world), and the description adds genuinely non-obvious behavior: the compact default view with content cut at ~300 chars and a trailing ellipsis, omitted bookkeeping/timestamp/scope fields, and the sub-threshold case returning an empty list with an explanation instead of weak matches. What it omits beyond annotations is the ranking method and any cost/latency characteristics, which keeps it short of a 5.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Purpose and routing are front-loaded in the first sentence and alternatives close the paragraph, with no filler phrasing. The long parenthetical describing the compact view is information-dense rather than padded, though it slightly interrupts the flow and duplicates schema text about full=true.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For an 18-parameter tool with no output schema, the description carries the important return-shape burden: compact field list, truncation marker, ranking by score, and the empty-with-explanation path. Pagination/ordering and expiration handling are left entirely to the already-complete schema, which is acceptable but leaves the description just short of exhaustive.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100% and each of the 18 parameters is documented in detail, including limit presets, graph-boost tail slots, tag AND/OR semantics and min_importance defaults. The description only re-elaborates the two most surprising ones (full=true and include_archived=true); for the remaining parameters it adds no meaning beyond the schema, so the baseline 3 applies.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

Opens with a specific verb+resource ('SEARCH memory by keywords') and narrows scope to 'a SPECIFIC piece of knowledge' with concrete examples ('API convention', 'license decision'). It explicitly distinguishes itself from recall_with_graph (follows links) and get_project_knowledge (pages one project's permanent knowledge), so an agent can choose without opening another schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

States the selecting condition directly: 'Use it to find one fact', and names two alternatives with the behavior that differentiates them (graph link-following vs. project knowledge paging). It also routes the agent to get_memory(key) for reading a single cut entry in full, giving a concrete when-to-use-this-instead rule.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.