Skip to main content
Glama

Exa Search

exa_search
Read-onlyIdempotent

Web search by meaning, not just keywords. Returns title, URL, published date and author per result; optionally page text, highlights, or one structured answer synthesised from the results. Search types: 'auto' (default, balanced), 'lite' (cheapest, fastest, no output_schema), 'deep-lite' and 'deep' (multi-step research). Categories: company, publication, news, personal site, financial report, people. max_age_hours controls freshness (0 = fetch live, -1 = cache only); historical snapshots are not available on this gateway's Exa plan. Billing note: results beyond 10 cost extra upstream. Example: exa_search({ query: "startups building AI agents for customer support", num_results: 10, type: "auto", category: "company" })

Example tool call

This call uses example arguments supplied in the tool’s input schema.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "type": "auto",
      "query": "recent research on protein folding with diffusion models",
      "num_results": 10
    },
    "name": "exa_search"
  }
}

Input Schema

NameRequiredDescriptionDefault
typeNoSearch type: 'auto' (balanced, default), 'lite' (low cost and latency), 'deep-lite' (light multi-step research), 'deep' (comprehensive multi-step research). Legacy neural/keyword/fast/instant/deep-reasoning are still accepted and mapped to the nearest current type with a warning.
queryYesThe search query, describe what you want by meaning, e.g. "recent research on protein folding with diffusion models"
_apiKeyNoOptional, your own Exa API key for higher limits; omit to use the shared Pipeworx key.
categoryNoOptional focus category. Documented values: company, publication, news, 'personal site', 'financial report', people. Other strings are sent to the vendor as hints with a warning; 'tweet' is retired and refused. 'research paper' maps to publication.
num_resultsNoNumber of results to return (default 10, max 100). Results beyond 10 cost extra upstream.
include_textNoIf true, include the clean parsed page text (up to 2000 chars) for each result. Default false.
max_age_hoursNoMaximum age of page content in hours, an integer from -1 to 720. 0 fetches live content (last 60 seconds); -1 uses only the cache. Omit for the vendor default (cache).
output_schemaNoOptional JSON schema (root type 'text' or 'object', at most 10 properties and 2 levels deep, every array with items). Returns one synthesised answer in `output` grounded in the results. Not supported with type 'lite'; adds about 2 seconds.
include_highlightsNoIf true, include the most query-relevant excerpts of each page. Default false (switched on automatically when max_age_hours is set without include_text).

Output Schema

NameRequiredDescriptionDefault
outputNoSynthesised answer when output_schema was given: a string or an object matching the schema
resultsYesMatching pages
warningsNoLegacy or undocumented argument values that were mapped or passed as hints
costDollarsNoUpstream cost breakdown in USD as reported by the vendor
search_typeYesSearch type actually used: auto, lite, deep-lite or deep
autopromptStringNoQuery rewrite, when the vendor returns one
upstream_warningsNoWarnings returned by the vendor

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed3 schema fields changed
    • changedInput schema / examples
      Previous value: -[
      -  {
      -    "num_results": 10,
      -    "query": "recent research on protein folding with diffusion models",
      -    "type": "auto"
      -  },
      -  {
      -    "category": "company",
      -    "include_text": true,
      -    "num_results": 5,
      -    "query": "startups building AI agents for customer support",
      -    "type": "lite"
      -  },
      -  {
      -    "category": "financial report",
      -    "include_highlights": true,
      -    "query": "Apple annual report 10-K",
      -    "type": "auto"
      -  },
      -  {
      -    "category": "news",
      -    "max_age_hours": 0,
      -    "query": "latest Federal Reserve interest rate decision",
      -    "type": "auto"
      -  },
      -  {
      -    "query": "OpenAI pricing page",
      -    "snapshot_as_of": "2025-01-15",
      -    "type": "auto"
      -  }
      -]New value: +[
      +  {
      +    "num_results": 10,
      +    "query": "recent research on protein folding with diffusion models",
      +    "type": "auto"
      +  },
      +  {
      +    "category": "company",
      +    "include_text": true,
      +    "num_results": 5,
      +    "query": "startups building AI agents for customer support",
      +    "type": "lite"
      +  },
      +  {
      +    "category": "financial report",
      +    "include_highlights": true,
      +    "query": "Apple annual report 10-K",
      +    "type": "auto"
      +  },
      +  {
      +    "category": "news",
      +    "max_age_hours": 0,
      +    "query": "latest Federal Reserve interest rate decision",
      +    "type": "auto"
      +  }
      +]
    • changedInput schema / properties / include_highlights / description
      Previous value: -"If true, include the most query-relevant excerpts of each page. Default false (switched on automatically when max_age_hours or snapshot_as_of is set without include_text)."New value: +"If true, include the most query-relevant excerpts of each page. Default false (switched on automatically when max_age_hours is set without include_text)."
    • removedInput schema / properties / snapshot_as_of
      Removed value: -{
      -  "description": "Return the newest stored version of each page as of this ISO 8601 date-time or YYYY-MM-DD date (midnight UTC). Served from stored versions only; the live page is never fetched.",
      -  "type": "string"
      -}
  2. Added

TDQS

A3.6/5.0
Behavior3/5

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

Annotations already declare readOnly/openWorld/idempotent/non-destructive, so the safety profile is covered. The description goes beyond them with genuinely useful operating facts: results beyond 10 cost extra upstream, historical snapshots are unavailable on this gateway's Exa plan, legacy type names are mapped with a warning, 'tweet' is retired and refused, and output_schema adds ~2s and is incompatible with 'lite'. Missing a note on error behavior and rate limits keeps it from 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?

Front-loaded with purpose and return shape, then modes, categories, freshness, billing and a concrete example — a logical order for an agent scanning it. Dense but each clause carries content; it loses a point for restating schema facts (defaults, enum lists) rather than leaving them to the structured fields.

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 a 9-parameter tool with nested output_schema, it covers the hard parts: output_schema constraints and its latency and lite incompatibility, freshness semantics, cost cliffs, and gateway restrictions. The one substantive gap is routing against sibling search/answer tools, which leaves an agent with no stated tie-breaker.

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 coverage is 100%, so every parameter, enum value, range and default is already documented in structured form. The description mostly restates those same facts (type tradeoffs, categories, max_age_hours values, billing beyond 10); the only genuine additions are the gateway-plan limitation and the worked example, which the schema mirrors in its examples array. Baseline 3 is correct.

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?

Opens with a specific verb+resource plus a differentiator: 'Web search by meaning, not just keywords', and enumerates the return fields (title, URL, published date, author). It never names a sibling such as exa_answer or search_within, so the agent must infer the boundary itself — especially since the output_schema mode ('one structured answer synthesised from the results') overlaps with an answer-style sibling.

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?

Gives real selection guidance among the tool's own modes: 'auto' balanced, 'lite' cheapest/fastest, 'deep-lite' and 'deep' for multi-step research, plus category choices and max_age_hours semantics (0 = live, -1 = cache only). It stops short of telling the agent when to use this instead of an alternative search/research tool, so mode selection is covered but tool selection is not.

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.