Skip to main content
Glama

Trillboards DOOH Advertising

semantic_search_observations

Search observations by semantic similarity. Find moments that match a description like "lunch rush at fast casual restaurants" using vector embeddings.

Uses 768-dimensional Gemini embeddings on observation payloads to find promoted observations matching a natural language query via approximate nearest-neighbour (ANN) cosine similarity search over a Lance IVF_PQ index.

CONSISTENCY: results are APPROXIMATE and EVENTUALLY CONSISTENT.

  • Approximate: retrieval is ANN, not an exhaustive scan (measured recall ~0.96 against exact KNN), so an identical query may omit a borderline match.

  • Eventually consistent: the index is served from a replicated pool whose replicas refresh independently, so for up to 5 minutes after new observations are published, two identical calls may return slightly different result sets. The difference is confined to the VISIBILITY of newly-published observations; the relative ranking of already-visible ones does not change. Do not use this tool where a repeatable, exhaustive result set is required.

TIME BOUND: searches the last 30 days by default. Pass filters.time_range to widen or narrow it; the window actually applied is echoed in metadata.time_range. Observations are retained for 90 days.

WHEN TO USE:

  • Finding observations that match a conceptual description

  • Discovering contextual moments across the screen network

  • Searching for audience situations ("families waiting in line", "professionals on coffee break")

  • Finding commerce patterns ("high purchase intent near checkout")

RETURNS:

  • data: Array of matching observations ranked by semantic similarity, each with:

    • observation_id, device_id, venue_type, observation_family

    • observed_at, payload, confidence, evidence_grade

    • similarity: Cosine similarity score (0-1, higher = more relevant)

  • metadata: { result_count, query_embedding_model, search_scope, time_range }

  • suggested_next_queries: Related semantic queries to explore

EXAMPLE: User: "Find lunch rush moments at fast casual restaurants" semantic_search_observations({ query: "lunch rush at fast casual restaurants with high foot traffic", filters: { venue_type: ["restaurant_qsr"] }, limit: 20 })

User: "Find moments with high emotional engagement" semantic_search_observations({ query: "audience showing strong positive emotional reactions", filters: { observation_family: ["audience"] }, limit: 10 })

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
limitNoMaximum results to return (default: 20, max: 100)
queryYesNatural language description of the observation moments to search for
filtersNoAdditional structured filters to narrow semantic search

TDQS

A4.6/5.0
Behavior5/5

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

With no annotations provided, the description carries the full burden and excels. It discloses approximate retrieval (ANN, recall ~0.96), eventual consistency (up to 5 minutes variance for new observations), time bounds (30-day default, 90-day retention), and explains the stability of rankings. This is rich behavioral detail well beyond a simple read-only hint.

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

Conciseness5/5

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

The description is long but logically organized with clear sections (CONSISTENCY, TIME BOUND, WHEN TO USE, RETURNS, EXAMPLE). It is front-loaded with the core purpose, and every section contributes meaningfully. The length is justified given the tool's complexity and the absence of an output schema.

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?

The tool is complex (semantic search, filters, nested objects, no output schema), and the description covers all necessary aspects: behavior, limitations, time bounds, return schema, and usage examples. It explicitly documents the response fields (data, metadata, suggested_next_queries) and the values within data, making it self-contained for an agent.

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

Parameters4/5

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

The input schema already has 100% coverage of parameter descriptions, so the baseline is 3. The description adds value with realistic examples showing how to phrase queries and filters (e.g., 'lunch rush at fast casual restaurants with high foot traffic'), clarifies default behavior for limit and time_range, and explains the returned similarity scores. This goes beyond the schema but does not cover edge cases like time range format.

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

Purpose4/5

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

The description clearly states the tool's function: 'Search observations by semantic similarity' and provides a concrete example query. The mechanism (vector embeddings, ANN) is specific. It does not explicitly differentiate from sibling tools like find_similar_moments or semantic_audience_search, so it meets 'clear but no sibling differentiation' rather than the full 5.

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?

The 'WHEN TO USE' section provides explicit scenarios (e.g., 'Finding observations that match a conceptual description'), and the description explicitly states 'Do not use this tool where a repeatable, exhaustive result set is required.' This covers both when and when-not, giving clear usage guidance even without naming alternative tools.

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

B3.3/5.0
Disambiguation2/5

There are exact duplicates (get_task_status/tasks_get, list_tasks/tasks_list) and several overlapping analytics, attribution, and semantic search clusters (get_attention_metrics vs get_creative_attention vs get_social_attention; find_similar_moments vs semantic_search_observations; get_campaign_attribution vs get_multi_touch_attribution vs get_roas). Detailed descriptions help, but with 83 tools an agent will frequently struggle to pick the right one.

Naming Consistency3/5

Most tools follow a snake_case verb_noun pattern (list_devices, create_campaign, delete_webhook), but there are notable inconsistencies: list_* and get_* are used interchangeably for list operations, attention tools mix conventions (get_attention_metrics vs get_creative_attention vs get_social_attention), and the legacy tasks_get/tasks_list names break the established get_task_status/list_tasks pattern.

Tool Count1/5

83 tools is an extreme count for a single MCP server, spanning device management, sensing, campaigns, media buys, attribution, webhooks, billing, API discovery, and AdCP protocol concerns. This is a broad API surface dump rather than a focused tool set, and it would be far better split into several coherent servers.

Completeness2/5

Despite the enormous surface, core campaign lifecycle is incomplete: create_campaign explicitly tells the agent to use update_campaign to activate a campaign, but no update_campaign tool exists, and there are no list/delete campaign tools. Significant capabilities exist for analytics, attribution, and webhooks, but the primary advertising workflow has a dead end.

Resources