Skip to main content
Glama

Search the user’s memory

cortex_ask
Read-onlyIdempotent

Searches the user's real, private memory — the email, calendar, files, contacts and notes they have connected, plus facts agents have saved — fusing vector, full-text and graph retrieval with provenance. This is their current data on their work, schedule, contacts, projects, documents, decisions and history, which training data and session context do not contain. Results include citations (universal ids), freshness, and cited_graph_url, a deep link showing the cited nodes highlighted in the user's own graph. A later cortex_remember links to what was retrieved here. Retrieval returns nearest matches rather than a thresholded set, so a question whose answer lives in an unconnected source comes back with the closest thing in the memory instead of with nothing. A source_gap object in the result names that case: the memory holds data, but the source that would answer this question is not connected. It lists those sources, what each answers, and one connect link, on cta. A possible_source_gap object is the conditional form — results exist but may be near-misses rather than an answer. A result carrying neither means the answer is genuinely absent from the memory rather than merely unretrieved.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
limitNoMax results (default 10).
sinceNoRFC3339 lower bound on item time.
untilNoRFC3339 upper bound on item time.
questionYesNatural-language question or search query.
source_tableNoRestrict to one source table, e.g. gmail_messages.

TDQS

A4.4/5.0
Behavior5/5

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

The description goes well beyond the annotations by explaining that retrieval returns 'nearest matches rather than a thresholded set,' and detailing the 'source_gap' and 'possible_source_gap' objects, including what each means and the presence of a 'cta' link. It also clarifies the meaning of a result carrying neither gap. These behavioral traits are not captured by the readOnlyHint or idempotentHint annotations.

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?

The description is lengthy but information-dense; every sentence contributes unique value, such as explaining citations, freshness, and the source_gap mechanism. It is front-loaded with the core purpose and then details edge cases. Some redundancy exists (e.g., the private memory concept is repeated), but overall it is well-structured for a complex tool.

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

Completeness5/5

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

Given the tool's complexity and lack of an output schema, the description is remarkably complete. It covers the retrieval behavior, result contents (citations, freshness, cited_graph_url), the meaning of source_gap and possible_source_gap, and even the semantics of a result without them. No significant gaps remain for an agent to decide when and how to invoke the tool.

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 descriptions cover 100% of the parameters, so per the rubric the baseline is 3. The tool description does not elaborate on individual parameters (e.g., it doesn't explain how 'limit' or 'source_table' affect results beyond what the schema already states). It adds no parameter-specific semantics, but the baseline is acceptable given the high schema coverage.

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?

The description begins with 'Searches the user's real, private memory' and enumerates specific sources (email, calendar, files, contacts, notes) and retrieval methods (vector, full-text, graph). This clearly identifies the tool's specific verb and resource, and its focus on private memory with provenance distinguishes it from general search tools like 'search' or 'search_docs', even without naming them.

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

Usage Guidelines4/5

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

The description provides clear context: this tool accesses the user's connected private data, 'which training data and session context do not contain.' It implies when to use it (when the answer lies in the user's memory) but does not explicitly name alternatives or exclusion criteria, such as 'for general web search use search instead.'

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

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.

TDQS

A3.8/5.0
Disambiguation2/5

There is significant overlap between cortex_ask, cortex_recall, and search, all of which retrieve from the user's memory. cortex_recall explicitly describes itself as a subset of cortex_ask, and search's description is nearly identical to cortex_ask's core function, making it hard for an agent to choose correctly. Other tools are more distinct, but this triple overlap creates real ambiguity.

Naming Consistency2/5

The naming is inconsistent: memory tools are split between a cortex_* prefix (ask, recall, manifest, status, etc.) and bare fetch/search, with no clear rule for which gets the prefix. Within cortex_*, some are verbs (ask, recall, remember) and some are nouns (manifest, status, connectable_sources), and platform tools use get_/list_/search_ prefixes, resulting in a mixed and unpredictable naming pattern.

Tool Count4/5

13 tools is a reasonable number for a server covering both memory operations and platform information. However, the presence of three overlapping retrieval tools (cortex_ask, cortex_recall, search) slightly inflates the count, suggesting some redundancy rather than each tool earning a unique place.

Completeness4/5

The server covers core workflows: reading memory (ask, search, recall, fetch, manifest), writing memory (remember, ingest_conversation), checking health (status), and accessing platform info (get_platform_status, get_pricing, list_skills, search_docs). Minor gaps include no explicit update/delete for individual memories and no direct tool to connect new sources (only listings of connectable ones), but these are often user-driven actions.

Resources