Gong

Use Gong for sales-call evidence. Call metadata alone is not enough for a deep dive that asks what happened in customer conversations.

Answer Questions With Gong's Own Search First

Gong's MCP (ask_account, ask_deal, generate_brief) searches and analyzes the calls server-side and answers in seconds. That is how Claude answers Gong questions, and it is the default here too.

• A question about an account, deal, customer, competitor, or topic ("what did X say about Y", risks, objections, themes, next steps, call prep) → run-gong-native-insight first, one consolidated call. Do not page /calls or /calls/extensive, and do not pull transcripts, to answer a question.
• Reach for the raw paths below only when the user needs verbatim quotes, an exact count, or proof that something was absent across all calls, or when the native operations are not connected (gong-native-insights lists them).
• If a blocking Gong request returns a size-limit error, stop paging and switch to the native operation or a narrower gong-calls search. Never retry the same wide window.

Actions

• gong-native-insights — lists the connected provider-native qualitative synthesis operations (ask_account, ask_deal, generate_brief) and their schemas. It never calls Gong AI or spends a credit request.
• run-gong-native-insight — sends one of those operations through Gong's connected MCP. This is the first choice for questions, themes, risks, summaries, and deck narrative when the native tools are connected. One invocation makes one Gong AI request (one credit request), so put related sub-questions in one call. It is on by default; a workspace owner can turn it off with configure-gong-native-insights, which never blocks the raw evidence path.
• account-deep-dive — first choice for a named account/deal when the answer needs HubSpot CRM context alongside Gong. For a question about what was said in the calls, use run-gong-native-insight first. It searches by account/deal/company/contact domain, loads Gong call details, and returns compact transcript excerpts for synthesis.
• gong-calls — bounded convenience reads: list recent calls, search by company/domain/person/email, fetch a single transcript by call ID, or return transcript excerpts for a small matching cohort. Its transcriptQuery is a local scan after batched transcript retrieval, not Gong's server-side keyword search.

For exhaustive, count, or absence-sensitive Gong work, use the raw provider API surface: use provider-api-catalog / provider-api-docs, then stage /calls/extensive or another raw endpoint with provider-api-request. Reduce staged tracker or call data with query-staged-dataset or a Data Program. Use provider-corpus-job only when raw transcript bodies are required and no native/indexed search can answer the question.

Route Synthesis and Evidence Separately

• Use run-gong-native-insight to answer the question where Gong retrieves and analyzes evidence internally. Its result is provider synthesis: coverage is unknown and raw transcript evidence is unavailable, so say that when the user asked for completeness.
• Use gong-calls or provider-corpus-job for quotes, exact counts, transcript records, coverage-sensitive claims, and proof that something was absent.
• Consolidate related native questions into one narrowly scoped request. Each native operation independently consumes a Gong credit request, so do not make one call per slide, bullet, or sub-question.
• From a sibling app, ask Analytics a natural-language question by default. Analytics owns Gong/HubSpot routing, source coverage, credentials, transcript strategy, joins, and synthesis. Use call-agent's direct action + input mode only when a trusted integration explicitly names one stable semantic read contract and supplies its complete bounded input. Never use direct actions as a workaround for a slow or failed delegated run.

Account Search Algorithm

Gong search pages POST /v2/calls/extensive directly with the requested date window and party data (not GET /v2/calls, which returns no parties). Each call is matched against both:

1. Title variants: the call title contains a generated search variant.
2. External parties: an external participant's name, email, or domain matches a generated search variant.

This catches calls titled "Builder <> Acme" when you search "Acme Corp" by matching the "acme" name variant or "@acme.com" domain variant. The lib generates variants by stripping deal suffixes (- New Deal, - Fusion), corporate suffixes (Group, Inc, Corp, LLC), and deriving first word and email domain — always including the raw original. Variants shorter than 3 chars are filtered. This is why a broad company search can find more calls than an exact-title search.

Customer-Voice Extraction (externalMonologues)

When analyzing customer sentiment, objections, or voice of the customer:

• Use getEnrichedTranscript / request enriched transcripts — it maps each monologue to speaker identity and affiliation.
• Only use externalMonologues for customer/prospect statements. Monologues with affiliation === "Internal" are your own team's speech — never surface these as "what the customer said."
• The externalMonologues field on an enriched transcript is the pre-filtered customer-voice signal; use it directly.

Patterns

For account or deal deep dives:

1. Ask run-gong-native-insight first for what customers said, risks, objections, and next steps. Call account-deep-dive first when the request also needs CRM context, contacts, stages, amount, close date, or an overall opportunity narrative, or when native operations are not connected.
2. Use gong-calls for targeted follow-up searches by account name, domain, person, or email.
3. Set includeTranscripts=true when the user asks for context, risks, objections, next steps, decision process, sentiment, or a "deep dive".
4. Use transcriptLimit around 3-5 for a first pass. For broad coverage of a named account, increase to 10-20 — the action supports up to 50. Increase when the returned calls don't cover the time window or key people you need. For a bounded "review all calls" request with at most 50 matches, make one call with exhaustive=true, after, before, and includeTranscripts=true; do not split discovery and transcripts into separate agent tool calls.
5. Use the compact transcript excerpts returned by includeTranscripts=true. Do not fetch raw individual transcripts unless the user asks for exhaustive quoting, debugging, or export.
6. Ground qualitative findings in the transcript excerpts and state how many calls were inspected.
7. For bounded account coverage, use cursor/offset only when the cohort is intentionally bounded. If the action reports truncation, treat coverage as incomplete and switch to tracker staging with provider-api-request plus query-staged-dataset or a Data Program, or to provider-corpus-job when raw transcript bodies are required. Do not increase the limit or keep paging gong-calls for broad scans. Do not use gong-calls(exhaustive=true) or one blocking provider-api-request call as the analysis. Stage the raw response and reduce it with a Data Program or query-staged-dataset; use provider-corpus-job for checkpointed transcript batches when raw bodies are necessary.

Quota and Coverage Rules

• Every Gong list or transcript page consumes provider quota. Budget pages and transcript batches before starting a broad scan; do not spend quota retrying the same bounded action after it reports incomplete coverage.
• A terminal HTTP 429 is terminal for that attempt. The shared provider transport may perform bounded cooldown/retry handling from Retry-After, but the agent must not add another retry loop. Report the rate-limit failure and switch to the staged tracker or checkpointed provider-corpus-job path only if that route is still appropriate.
• Exhaustive, count, and absence claims require complete coverage of the stated cohort. If any page, transcript batch, or provider request fails or is truncated, report the gap and do not make the claim.

Example:

gong-calls(company: "The Knot", days: 180, limit: 20, includeTranscripts: true, transcriptLimit: 10)

Gong search is best-effort: it matches title plus external participant names, emails, and domains through /calls/extensive. Treat call details and transcript excerpts as evidence; treat missing coverage as a gap, not proof that the topic never came up.

Title/party matching is discovery only. Never use a title pre-filter as proof that a phrase is absent from transcripts.

If transcript loading fails for a call, report that gap instead of inferring the conversation content from title, date, or participants.

When a single transcript is needed, gong-calls(transcript: "...") returns compact extracted text by default. Set rawTranscript=true only for debugging/export, and never pass raw transcript payloads into save-analysis.

Complete-Coverage Transcript Scan

When the question is "do ANY of these calls mention X?" or "how many calls across this cohort mention X?" — where missing a single call makes the answer wrong — do NOT conclude absence from a sampled includeTranscripts result. Exhaustive claims require complete coverage of every call in the stated cohort. Choose the smallest complete path for the bounded cohort:

1. Configured keyword tracker. If the requested term is already a Gong keyword tracker exposed in API results, use provider-api-request against /calls/extensive with contentSelector.exposedFields.content.trackers. Stage the paginated calls, then flatten and filter tracker ids, names, or phrases with query-staged-dataset or a Data Program. This is the preferred broad keyword path because Gong has already indexed the tracker. It does not answer arbitrary terms that are not configured as trackers.

2. Named account, up to 50 calls, qualitative review. Make one gong-calls call with exhaustive=true, a bounded after/before window, and includeTranscripts=true. The action discovers every matching call and fetches transcript excerpts in batches, returning transcriptCoverage. Do not fetch the same call IDs one at a time afterward.

3. Bounded arbitrary mention/absence search, up to 200 calls. Make one gong-calls call with exhaustive=true, the date window, transcriptQuery, and an adequate transcriptScanLimit. This is a bounded local scan after batched transcript retrieval. Report its coverage fields and only make an absence claim when coverage is complete.

4. Larger or multi-account arbitrary corpus. Use provider-api-catalog(provider: "gong") and its corpusRecipes if you need the exact shape. Stage call ids with the raw API, then use provider-corpus-job with mode: "batch-search" and 20-call transcript batches. The canonical request is POST /calls/transcript with batch.itemBodyPath: "filter.callIds", batch.responseItemsPath: "callTranscripts", batch.batchSize: 20, search.textPaths: ["transcript"], and search.idPaths: ["callId"]. Feed the staged/discovered call ids through batch.inputDatasetId + batch.inputValuePath or through batch.items. This is a durable full scan, not a server-side keyword query; do not implement it as a per-call loop in run-code or a delegated agent.

When delegating a corpus chunk to an Agent Teams sub-agent, include the exact staged dataset id or scratch/... path and the output/checkpoint path in the task brief. The child starts with a clean context and must read the staged input before making any provider request. If that input is unavailable, it must report the missing input rather than re-fetching the entire Gong corpus.

5. Use run-code only for joins/reductions around the corpus path. After the transcript job exists, use run-code, query-staged-dataset, or job results to join hits back to deals/accounts, compute variants, dedupe, and format evidence. A run-code loop over gong-calls(transcript: id) is a fallback for small or awkward sets, not the default for broad scans.

Report coverage explicitly: deals in cohort, calls discovered, calls scanned, and matches found. Never turn "I inspected a sample" into "no call mentions X".

Limits (Current)

Parameter	Default	Max
limit (calls returned)	8	200
transcriptLimit	3	50
transcriptMaxChars	8 000	100 000

For large account deep-dives or competitive analyses spanning many calls, use limit: 50-200 and transcriptLimit: 20-50. For very large datasets, use the checkpointed provider-corpus-job path rather than paging the full Gong call list inside one agent action.