Njord MCP

Overview

Connect an AI client to read workspace records, Audiences, reports, campaigns, website activity, Signals and meeting notes. Read/write profiles add Audience management plus account/stakeholder changes. Both profiles include Njord AI conversations and usage guidance.

Endpoint: Copy the server URL shown when creating an access token in Settings > API Tokens.

Package: Njord DOE

Setup

Create an MCP token in Settings -> API Tokens for the space you want to connect. The token acts on behalf of the person who creates it. Choose read-only access unless your workflow needs the supported write tools.

What your client needs

A remote MCP client that supports HTTP POST requests to the environment's /mcp endpoint. No separate event-stream connection (SSE) is needed.
The MCP token in Authorization: Bearer <token>, or the supported X-MCP-Token fallback header
A compatible client may also use the URL fallback /mcp?token=<URL-encoded-token>, but prefer headers because URL credentials can leak into history, proxy/access logs and copied links
Check tools/list for the tools available to your token, their exact input/output fields, the connected space and public model/reasoning options. MCP, API and Webhook tokens are not interchangeable.

Steps

Open Settings -> API Tokens in the space you want the client to access.
Click Generate token. The Profile selector offers separate API and MCP profiles; choose an MCP profile for this connection.
Confirm the Space ID. The token can access only this space and stops working if its owner loses access to the space.
Choose MCP Space Read Only for direct reads, or MCP Space Read Write to manage Audiences, create/update accounts and stakeholders and delete stakeholders. Njord AI conversations follow their own permissions, explained below.
Get and copy the MCP server URL shown when creating the access token in Settings > API Tokens. Use that exact URL to configure your MCP client.
Set the token expiry if needed. The default is one month and the maximum lifetime is one year; leaving it unspecified does not create a non-expiring token.
Click Generate. Existing read-only tokens do not need replacing for reads; a new read/write-profile token is needed for direct writes.
Copy the full secret from the success dialog immediately and store it in a secret manager or the client's protected credential store.
Configure a remote HTTP MCP connection with the endpoint and Authorization: Bearer <token>. If custom Authorization is unavailable, use X-MCP-Token: <token>.
Connect or reconnect, initialize, then discover tools. Use xNjord.authoritativeSpaceId from tools/list for space-scoped inputs and get_njord_guidance for task-specific instructions.

Only generation and rotation return the full token. Listing returns a key preview and metadata, not a recoverable secret. Never paste a token into a prompt, shared URL, screenshot or source control.

Available tools

Targeting Audience

Seven Audience read tools and eight read/write management tools. Account techniques belong to Audiences. Inspect saved progress and partial results; accepted work is not necessarily complete.

get_audience_operation: Read uncached operation progress, committed counts and failure messages. QUEUED, RUNNING and CANCEL_REQUESTED are incomplete. SUCCEEDED confirms completion; FAILED or CANCELLED may retain committed changes. Refresh saved records before deciding what remains.
retry_audience_operation: Resume failed Audience work from its saved checkpoint. Requires read/write access; inspect partial results and newer conflicting changes first.
cancel_audience_operation: Request cancellation of membership work at a committed batch boundary. Partial changes remain. Poll until CANCELLED; status/archive reconciliation cannot be cancelled.
get_audience: Read one Audience, including type, status and selected account techniques.
list_audiences: Page through Audiences using optional type, status and name filters. Pagination belongs inside filter; metadata.count is the full matching total.
get_audience_insights: Read Audience counts and supported performance breakdowns. Use the live tool schema, then inspect returned cohort, period, metric basis, coverage and notes.
list_audience_members: Read the selected Audience's members with supported selection, sorting and pagination. Performance views can have a different cohort; use only options present in the live tool schema.
list_entity_audience_memberships: Read current Audience references for an account or stakeholder. Account membership does not enroll its stakeholders.
list_audience_audits: Read dated Audience action history, actor, status and technique changes.
create_audience: Create a Draft ACCOUNT or STAKEHOLDER Audience. Account creation requires one or more techniques: LINKEDIN, IP, GEO, RETARGETING. Stakeholders use Individual Targeting; omit techniques. Requires read/write access.
update_audience: Edit name, description or account techniques. Every account update includes a nonempty techniques selection. Active technique changes return operationId; poll progress. Requires read/write access.
set_audience_status: Set IN_USE to enable targeting or INACTIVE to retain members without targeting. Poll returned progress. Draft is initial only. Requires read/write access.
archive_audience: Archive a Draft or Inactive Audience. Populated archival retains members and join dates and becomes terminal/read-only. Empty archival deletes the empty Audience immediately; history remains. In-use must first become inactive; Legacy Audiences cannot be archived. Requires read/write access.
clone_audience: Clone current members into a new Draft Audience; wait for completion before changing the clone. Requires read/write access.
change_audience_members: INCLUDE or EXCLUDE a snapshot selected by exactly one source: ids, all:true, accountFilter or stakeholderFilter. Use excludedIds for exceptions, including unloaded results. Account membership never enrolls stakeholders. Poll the returned operation and inspect saved counts; other active Audiences can keep excluded entities targeted. Requires read/write access.

Reports

Start with list_report_catalog and get_njord_guidance(topic=reporting). Use only live catalog entries marked available and their window, sort and metric-filter contracts; APIFP and DPIFP are not advertised in the public MCP report catalog.

list_report_catalog: Discover available report types, supported windows, metric-filter keys and sort keys. top is an activity-positive shortcut, not neutral pagination: media reports require clicks > 0 and website reports require visits > 0.
run_report: Run a known report using its discovered schema, including type.type, filter.space, duration/dateRange and supported sort. Use filter.take for neutral pagination; filter.top changes row eligibility as well as limiting rows.
report_accounts: Run compact account/company report rankings with supported filters, metric thresholds and count mode. Preserve identical predicates when switching between exact counts and matching rows.
report_stakeholders: Run stakeholder performance or impact-movement reports. Use ITMP with inTimeSpan metrics for bounded activity; use ITMIM only for changes to Reached, Aware or Engaged, not the full reached population.

Accounts and stakeholders

Use entity tools for detail records and actual rosters. Follow each tool's strict schema, keep space scope bound to the credential, and read metadata.count rather than a page length for supported full-population counts.

get_account: Fetch one account by id, including lifecycle-dependent health/RMF and an optional provider-neutral crm mapping array when available. Use list/report project options for additional compact roster fields.
list_accounts: List/filter accounts. Put take, skip, archived and includeArchived inside filter, never at the top level. Active accounts are the default; filter.archived=true selects archived records and filter.includeArchived=true includes both.
list_account_facets: Get account distributions and counts by supported categories such as owner, industry, account type or market segment. Use the discovered category/filter contract.
get_space_targeting_summary: Get the Space Targeting Summary: account/stakeholder targeting counts and supported segmentation, status, type and country distributions.
get_stakeholder: Fetch one stakeholder by id, including optional provider-neutral crm mapping references when available. Use list/report project options for optional roster classification and relationship fields.
list_stakeholders: List the active stakeholder population and filter by supported account, deal, owner and other predicates. metadata.count is the matching roster count; an ITMP report is not a substitute because its default activity selection can exclude inactive stakeholders.

Deals

Read deal records and pipeline context using the tools available to your token.

get_deal: Read one deal by its returned record ID, including the pipeline context exposed by discovery.
list_deals: Find matching deals with the discovered filters and sort. Page with filter.take/filter.skip and use metadata.count for the full matching count.

Account and stakeholder writes

Requires MCP Space Read Write. Each tool accepts an array, even for one record. Changes are applied immediately, with no server-side preview or extra approval step. A batch can partly succeed; check each result before retrying.

create_accounts: Send spaceId and input:[{name,website,...}]. Both name and website must be nonblank; website uses existing normalization and domain is derived, not a separate input. Follow-up IDs are returned in structuredContent.items[].id.
update_accounts: Send input:[{id,...changes}] with at least one changed field per row. Omitted fields stay unchanged; scalars replace and arrays replace the whole array ([] clears). Name/website cannot be cleared. Omit website on unrelated partial updates; changing it rederives domain.
create_stakeholders: Send spaceId and input:[{accountId,firstName,lastName,...}]. The account must exist in this space and all three identity fields are required. Email, phone and title are optional. Targeting uses Audience membership. Use structuredContent.items[].id for subsequent calls.
update_stakeholders: Send input:[{id,...changes}]. Omit unchanged fields; provided arrays replace, never append. First/last names cannot be cleared. Optional contact/text fields can be cleared with an empty string. Account/deal reassignment is not supported by this tool.
delete_stakeholders: Send ids:[...] with 1 to 1,000 stakeholder IDs from this space. Deletion is immediate and permanent, including linked events, signals and Individual Targeting reports. Inspect deletedIds, counts and failures; account deletion is not exposed.

Campaigns and ads

Use campaign-level tools for inventory and campaign/ad performance, not as replacements for account or stakeholder reports. LinkedIn supports its documented windows; Individual Targeting campaign metrics are cumulative and reject bounded date ranges.

list_campaigns: Browse campaigns using the supported platform, ID, status, window and sort inputs. For supported LinkedIn windows, an explicit dateRange takes precedence over duration.
list_campaign_ads: Inspect ad inventory and status inside a selected campaign, using the identifiers returned by campaign discovery.
get_campaign_ad: Fetch a specific ad using the platform and ad identifier required by its input schema.
list_campaign_performance: Rank campaign-level clicks, impressions and other supported metrics. Do not present a campaign ranking as an account media summary or stakeholder performance report.

Website behavior

Website tracking tools describe observed company/lead traffic, not audit activity or meeting history. Discover companies first, then inspect a known lead's visits or an account's pages.

list_website_tracking_accounts: List tracked website companies/accounts with the supported scope and time filters. Use returned lead identifiers for visit evidence instead of guessing an account ID is a lead ID.
list_website_visits: Return visit-timeline evidence for one known website lead ID using the discovered filters and pagination.
list_website_top_pages: Show page-level behavior for one known account, including supported attached source/visitor evidence. This is account-specific, not a space-wide top-pages report.
get_website_tracking_insights: Get aggregate identified traffic, daily visits, countries, industries and traffic sources. Supported scopes include TARGETED_ACCOUNTS, NON_TARGETED_ACCOUNTS and ALL_ACCOUNTS.

Signals and meetings

Prefer Signals V2 for current signal discovery and pagination. Meeting tools provide permission-aware, safe note summaries and search evidence, not raw transcript export.

list_signals_v2: Preferred signal tool for timelines, filters, pagination and supported sales insights. Follow its discovered schema rather than carrying over the legacy array contract.
list_signals: Deprecated legacy signal tool retained for existing clients. New workflows should use list_signals_v2; legacy arrays are wrapped in structuredContent.items for MCP compatibility.
list_meeting_notes: List safe meeting-note summaries and supported analysis/score fields. Use only returned evidence; do not assume full notes or transcripts are available.
search_meeting_notes: Search meeting-note content for subjects such as objections, procurement or next steps, within the supported privacy scope and the token owner's access.

Njord AI and guidance

Both profiles include Njord AI conversations and usage guidance. A read-only token limits the direct tools; it does not prevent changes that the token owner's usual Njord AI permissions allow.

ask_njord_ai: Send text and a modelId from the public model catalog. Optionally choose a supported modelOptions.reasoningEffort. Leave out chatId to start a chat, or reuse the returned chatId to continue as the token owner. Events are returned together in structuredContent.events, not streamed live to the MCP client; check for completion or errors. Only conversation, model and reasoning inputs are supported, not shortcut or signal-analysis modes.
get_njord_guidance: Read task-specific guidance for overview, reporting, accounts, deals, stakeholders, campaigns, website, notes_and_signals or conversation. This exposes curated usage rules, not internal prompts or private resolver names.

Reports

Code	Report	Notes
`AMPS`	Account media summary	Media Summary. ALL_TIME is cumulative; bounded LinkedIn/Individual Targeting metrics use the selected window, while IP/GEO remain cumulative, so overall intentionally mixes bases. Website named windows use the viewer-local calendar.
`SMP`	Space Media Performance	One row per selected technique; no cross-technique grand total. LinkedIn/Individual Targeting support bounded windows; IP/GEO stay cumulative and non-comparable. Account-cohort filters narrow the scope using current relationships.
`SMPC`	Space Media Period Comparison	Compare a bounded period with the immediately preceding equal-length period. LinkedIn/Individual Targeting are comparable; IP/GEO historical comparison is unavailable. Respect comparability/availability fields.
`STAM`	Space Top Account Movers	Rank gains/declines in clicks or impressions for LinkedIn or Individual Targeting over adjacent equal-length periods. Program/cohort selection uses current account type, not a deal-program field.
`ALIP`	Account LinkedIn performance	LinkedIn-only. Bounded metrics use per-account/provider current and pre-window snapshots; ALL_TIME uses latest cumulative values. top selects only click-positive rows, not the whole account roster.
`AIPP`	Account IP targeting performance	ALL_TIME-only latest cumulative IP metrics. Bounded duration/dateRange is rejected; do not silently substitute another channel. top requires clicks > 0.
`AGEOP`	Account geo targeting performance	ALL_TIME-only latest cumulative GEO metrics. Bounded duration/dateRange is rejected; do not present cumulative values as period activity. top requires clicks > 0.
`AITP`	Account individual targeting performance	Account-level Individual Targeting rollup: bounded-window deltas or ALL_TIME cumulative values. This is one channel, not all media. top requires clicks > 0.
`ITMP`	Stakeholder individual-targeting performance	Use inTimeSpan.* for bounded activity and overall.* for cumulative context. For everyone reached in a period, use inTimeSpan.impressions gt 0 with resultMode=count. top selects only stakeholders with clicks > 0; default report eligibility is not the full stakeholder roster.
`ITMIM`	Stakeholder media impact movement	Find transitions using filter.stakeholderImpact=REACHED, AWARE or ENGAGED in the selected period. Movement is not total reach: use ITMP impression thresholds for all stakeholders reached during that period.
`AWP`	Targeted-account website performance	ALL_TIME uses cumulative visits/pageViews; bounded windows use spanVisits.*. top is visit-positive (visits > 0), unlike media reports' click-positive rule.
`NTAWP`	Non-targeted website company/lead performance	Rows have no mapped targeted account. ALL_TIME uses cumulative visits/pageViews; bounded windows use spanVisits.*. top requires visits > 0.
`NTAAR`	Non-targeted LinkedIn organization performance	LinkedIn-only rows with no mapped account. Bounded metrics use per-organization/provider snapshot deltas; ALL_TIME is cumulative. top requires clicks > 0.

Operations and limits

Requests are limited by client IP before sign-in and by token after sign-in. Defaults are 200/minute and 2,000/hour per IP, plus 100/minute and 1,000/hour per token, with a 120-second temporary block. Wait for the time shown in Retry-After and reduce request volume. After a write times out, check whether it succeeded before sending it again.

POST /mcp
200/minute, 2,000/hour
100/minute, 1,000/hour
120 seconds

Review

Review tokens in the space's API Tokens list. It shows a safe preview and details, never the full secret. Each space supports up to five unexpired MCP tokens shared across its MCP profiles; use a separate token for each client or workflow where possible.

Review the key ID/preview, MCP type, profile, issuing user, expiry, last-used and last-rotated timestamps
Existing MCP_SPACE_READ_ONLY keys continue to read; issue MCP_SPACE_READ_WRITE for direct writes
For troubleshooting, MCP call logs identify the space, token key ID, token owner, client, tool, response time and outcome.
Call logs include authenticated requests and responses and are retained for 30 days. Cleanup runs automatically and may occur after that period. Treat these logs as business data.
Read, write and authenticated rejection audits use the MCP log store. Secret-shaped values are redacted; a log-store failure emits a warning and preserves the operation result. API requests have a separate log store.

Rotate

The token owner can rotate it to replace the secret without changing its access profile. To move from read-only to read/write, create a token with the new profile.

The old secret stops authorizing new requests; copy the new full token immediately
Update the client's secret store and reconnect so the client rediscovers its permitted tools
Rotate after suspected exposure or on your normal security schedule; issue a separate token for a different owner or access profile

Revoke

Revoke a token from its space when access is no longer needed. This permanently deletes the credential and its stored secret.

The revoked token cannot authorize further requests; create a new credential if access is needed later
Revocation is space-scoped and does not delete the business records already created by the token
Call logs keep their separate 30-day retention period. Revoking a token does not remove those logs or undo work already started.

Expire

Every issued token has a bounded lifetime: one month by default and no more than one year.

Choose shorter lifetimes for experiments, partners and temporary integrations
Review expiry before a scheduled workflow depends on the token
An unspecified expiry uses the default; it does not mean unlimited access

Safety checklist

Use a separate token per client or workflow where possible, and choose only the access it needs.
Keep tokens in headers and protected storage; URL-token fallback is a last resort.
Copy a new or rotated secret once, then retain only a safe preview in shared operational notes.
Respect the single-space scope and the issuing user's permissions; use discovery rather than guessing a Space ID.
The read-only profile restricts direct tools, not Njord AI conversations. ask_njord_ai acts as the token owner and may make changes if that user's configured Njord AI permissions allow them.
Direct write calls apply immediately. Client-side confirmation may still be appropriate, especially before permanent stakeholder deletion.
Treat arrays in updates as replacements, inspect partial failures, and never blindly repeat a timed-out create/delete batch.
Rotate exposed tokens, revoke unused ones, review expiry dates and remember that call logs may contain business data.

Things to know

Check the tools available to your token

Read profiles expose 33 tools; read/write exposes 46. Audience adds seven reads and eight management tools. Reconnect after changing keys and follow tools/list schemas and authoritativeSpaceId.

Read-only tools and Njord AI have different permissions

MCP_SPACE_READ_ONLY blocks the direct create, update and delete tools. By design, ask_njord_ai uses the token owner's usual Njord AI permissions instead. A conversation may make changes when those permissions allow it.

Required fields and website handling

Account creation needs a nonblank name and website. Njord checks the website and derives the domain automatically. Stakeholder creation needs an existing accountId in the same space plus firstName and lastName. Updating other fields on an older account does not require adding a missing website.

Updates replace lists; they do not add to them

Each update needs id and at least one changed field. Leave fields out to keep their values. A supplied list replaces the whole list, and [] clears it. To add an item, read the current list and send the complete new list; another user's edit could occur in between. Names and account website cannot be cleared. These tools cannot move a stakeholder to another account or change its deal assignments.

Targeting uses Audience membership

Create entities, then add them explicitly to an Audience. Member join dates belong to Audience reads. Entity writes do not accept retired targeting flags, dates or actors. Account and stakeholder memberships are independent; only in-use Audiences contribute to effective targeting.

Techniques belong to account Audiences

Choose one or more of LINKEDIN, IP, GEO and RETARGETING on account Audience creation and every update, including name-only edits. Stakeholder Audiences use Individual Targeting. Poll active technique reconciliation before claiming effective targeting changed.

Finding and archiving accounts

list_accounts uses filter.archived/includeArchived and filter.take/skip. Reads expose archived rather than verified. Account archival closes its Audience memberships. Audience archival has different semantics: populated Audiences retain members and join dates, contribute no targeting and become read-only; empty archival deletes the empty Audience.

top changes which rows qualify

For media-performance Top-N shortcuts, top includes only rows with clicks > 0; for website reports, it includes only rows with visits > 0. It is not a neutral limit and must not answer total-population questions. Use take for pagination and the live catalog for report-specific behavior.

Counts, rankings and changes answer different questions

metadata.count is not the current page length. Sorting does not impose a threshold. Use supported metricFilters and resultMode=count for exact report thresholds, and repeat those predicates for rows. ITMP is activity-based; list_stakeholders gives the active roster. ITMIM finds impact transitions, not everyone reached during a period.

Check what each reporting period includes

Explicit report ranges use an inclusive start and exclusive end. AIPP/AGEOP are cumulative-only; bounded AMPS intentionally mixes channel bases. Check windowBehavior and comparability rather than labeling every metric a period total. Do not silently replace an unsupported technique with a different channel.

Request extra fields only when you need them

Only list_accounts, list_stakeholders, run_report, report_accounts and report_stakeholders accept the optional top-level project selector. Account values are crm/targetingActivated; stakeholder values are crm/type/trafficLight. Mixed reports apply selections to matching entity types. Detail tools can return optional provider-neutral CRM references without this selector. Compact reports retain available stakeholder email/linkedIn and account owners as id/name references, not full user records; internal storage fields, logoUrl and report-row discriminator details are excluded.

Results support structured and text clients

structuredContent always has an object root: native objects stay objects, arrays become {items:[...]}, and scalar/null values become {value:...}. content includes a text block serializing the underlying payload; for arrays it need not repeat the wrapper. Schemas are self-contained. Large Njord AI results may use a short text fallback directing clients to structuredContent.events. Public validation/tool failures are sanitized rather than exposing internal resolver/provider errors; inspect isError and do not treat a successful HTTP status alone as a successful call.

A timeout does not mean a change was undone

Records are saved independently. Creation returns record IDs; updates return counts, failures and a small updatedSample; deletion returns counts, deletedIds and failures. If the connection times out, check current records and any partial results before retrying.

Deletion is permanent

delete_stakeholders removes selected stakeholders plus their linked events, signals and Individual Targeting reports. There is no MCP dry-run or separate server approval step; your client may still request confirmation. Account deletion and deal writes are not exposed.

Find out which token made a call

Authenticated MCP reads, writes and rejected calls are recorded with workspace/key/owner, tool, correlation, request and response for 30 days. Secret-shaped values are redacted. Responses wait for an audit write attempt; a log-store failure preserves the operation result and emits a warning. Related activity uses source=mcp and retains automatic-rule provenance in triggerSource. Public API has separate request logs and source=public_api.

Njord AI responses arrive as a set of events

ask_njord_ai returns the conversation's events together rather than streaming them live to the MCP client. Check completion/error events and any partial response. A disconnected client, a period without new events or a configured response limit can stop the turn; review the result before retrying.

Meeting summaries are not transcript exports

Meeting tools return permission-aware summaries/search evidence. Website tools return tracked traffic, and Signals V2 returns signal evidence; none should be substituted for internal audit logs or assumed to expose raw recordings/transcripts.

Authentication or access is rejected

Check the endpoint, MCP credential type, expiry and current workspace access of the token owner. API and MCP credentials are separate. Copy the endpoint from Settings > API Tokens.

A tool is missing or an argument is rejected

Reconnect after changing credentials and refresh tools/list. Read-only profiles omit direct write tools. Follow the current input schema and nesting; a field visible in the app is not automatically an MCP argument.

Requests are rate limited

Wait for Retry-After, then reduce request volume. The same key shares its allowance across tools; changing tools does not reset it.

A request timed out or returned partial results

Check returned events, errors, record IDs and any saved operation progress before retrying. A timeout does not prove that a write was undone. Retry only confirmed remaining work.

An Audience change has not finished

QUEUED, RUNNING and CANCEL_REQUESTED are pending. Review committed counts and failure feedback. Use the permitted recovery action and refresh the roster; do not replay a whole request blindly.

A result has missing or unexpected metrics

Check the report/tool catalog, returned period, cohort and metric basis. Missing or unavailable values are not zero, and historical performance can differ from the current management roster.

Brand and Context controls are in the app

This MCP catalog has no direct tools for Brand uploads, file versions, colors, font links or extraction approval, or for Context Knowledge, Facts, Assets, Target Market, AI Instructions or private Memory. Use Brand and Context in the app and their Knowledge Base guides. Website tracking is traffic evidence, not website review; ask_njord_ai follows the connected user's context and permissions.

Historical Audience performance is a different roster

MEMBERSHIP can include former members and their historical intervals; OVERALL uses current members. LinkedIn/Individual Targeting, IP/GEO and Retargeting have different measurement bases. Read coverage and technique notes, keep unavailable values distinct from zero, and use a single supported account technique for comparable rankings. Follow the live schema for the exact arguments.

Saved Audience configuration can still need recovery

Membership counts and configured techniques can be saved while targeting reconciliation has failed. Surface the returned failure warning and referenced operation, inspect its feedback and use the permitted recovery action. Refresh effective targeting before reporting completion.

Example prompts

List available reports and explain which support a last-30-days window before running one.
Count stakeholders with at least one impression in the last 90 days, then show matching rows using the same filters.
Which stakeholders became Engaged in the last 7 days, rather than everyone already Engaged?
Show the top targeted accounts by clicks and disclose which channel metrics are cumulative.
List active accounts and deals, including the full matching count rather than just this page.
Rank our LinkedIn campaigns by clicks using a supported bounded window.
Show non-targeted companies visiting the website this month, then inspect visits for a returned lead.
Show recent Signals V2 evidence and search meeting notes for procurement objections.
With write access, create Example Company with website https://example.com, then create Alex Example linked to the returned account ID.
Add this stakeholder to an Individual Targeting Audience and activate it after membership processing completes.
Update these accounts, preserving all omitted fields and explicitly replacing the selected array values.
With write access, delete only these stakeholder IDs after explaining the permanent linked-data removal.
Continue my Njord AI conversation using its returned chatId and a supported model/reasoning setting.
List the deals matching this account and owner, keep the supported sort while paging, and report the full matching count.
With write access, include all currently matching accounts in this Audience except these IDs, then track the operation and report any saved partial changes.
Explain each Audience technique's returned performance basis and unavailable metrics; distinguish historical membership estimates from lifetime activity.
Compare this Audience's historical membership performance with its current members, show each technique's metric basis, and identify any reconciliation failures.

Strongest for

Space-scoped account, stakeholder and deal lookup
Metric-filtered account/stakeholder rankings and accurate matching counts
Channel-aware campaign, website and media-performance analysis
Signals V2 and permission-aware meeting-note evidence
Controlled account/stakeholder creation, updates and stakeholder deletion with a read/write key
Audience membership, explicit targeting techniques, retained archive rosters and operation recovery
Njord AI conversation continuity plus curated reporting/tool guidance

Not designed for

Account deletion, direct deal mutation or workspace/user administration
Cross-space access, arbitrary resolvers or unsupported report/window combinations
Treating Top-N rankings as the complete population or cumulative metrics as period deltas
All-or-nothing batches, preview-only writes or adding to lists without sending their complete new values
Assuming a read-only token also prevents every change through Njord AI
Raw transcript exports or unlimited, live-streamed MCP conversations