--- name: abm-dev description: Enrich B2B people and companies with abm.dev — verified work emails, LinkedIn profiles, firmographics, and AI research with a citation and confidence on every field. Use when the user needs contact data, company intelligence, email-finding, people search at target accounts, or CRM enrichment writeback. --- # abm.dev — B2B enrichment for agents abm.dev turns an email, a name + company, or a LinkedIn URL into ~89 canonical fields per record (43 person + 46 company), each with a source citation and a confidence score, synthesised from ~10 data sources — and writes the results back to your CRM. ## Connect - **MCP (preferred):** `https://mcp.abm.dev/mcp` — OAuth sign-in on first use. One-click from Claude: https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=abm.dev&connectorUrl=https%3A%2F%2Fmcp.abm.dev%2Fmcp - **REST API:** `https://api.abm.dev` with the `X-API-Key` header — mint a key at https://abm.dev/settings/api-keys. Same verbs: `POST /v2/enrich`, `POST /v2/search`, `POST /v2/source`. - **CLI:** `npx -y @abmdev/cli@latest init --all -k ` ## The core loop (match → enrich → write back) 1. **Match** the record in the CRM: `search_companies` / `search_contacts`. Filters are EXACT-match — match on `domain`/`email`, not display names ("Saffery" will not match "Saffery LLP"). No match? `create_company` / `create_contact`. 2. **Enrich** with the CRM id threaded through: `create_enrichment { type, input: { company_name, domain, crm_id } }` → returns a `job_id` (async, commonly 3–8 min). Add `refresh: true` to force a from-scratch run when the record was enriched recently. Poll `get_enrichment` every 30–60s until `completed` (a processing status with 0 fields is normal — do not give up before ~10 min), then read values with `get_enrichment_fields` and provenance with `get_enrichment_sources`. 3. **Write back**: `preview_writeback` shows the per-field keep/use diff; then `approve_enrichment` (REAL CRM mutation). If the preview shows no/wrong matched record, pass `crm_entity_id` on approve. ## Finding people (async search) `search { type: "people", company: "…", limit: 20 }` returns a `job_id` immediately — the multi-page walk takes a minute or two. Poll `get_search_job { job_id }` every ~20s until `summary.status === "completed"`; `data[]` grows page-by-page. Role-targeted alternative: `source` + `get_source` (the people-finder). Both act as the org's connected LinkedIn account and respect its daily quotas — spread large sweeps out. ## Ground rules - Every field is grounded in a fetched source and returned with a citation + confidence — never fabricate a value; if a fact is not in the result, say so. - Respect tool annotations: `approve_enrichment` and the CRM create/update tools are REAL mutations; preview before you write. ## Tool reference (48 tools) Tool visibility is gated per org/role feature flags — not every tool below is always present. ### Enrich - `create_enrichment` (WRITE) — Start an async enrichment of a person or company (type + input). - `get_enrichment` (read) — Poll an enrichment job (pass wait: true to LONG-POLL — the server holds ~25s and returns at the first change, replacing rapid polling) by its sequential id for current status: queued | processing | completed | failed … - `list_enrichments` (read) — List enrichment jobs for the current org, newest first. - `cancel_enrichment` (WRITE) — Cancel an in-flight enrichment job by id. - `get_enrichment_sources` (read) — Get per-source attribution results for an enrichment — which fields came from which source (professional profiles vs verified web sources). - `get_enrichment_fields` (read) — Return the ENRICHED RESULT VALUES for a COMPLETED enrichment job: each field with its value, confidence, and the winning source. ### Write back to CRM - `approve_enrichment` (WRITE) — Approve a completed enrichment and write its fields back to the connected CRM. - `preview_writeback` (read) — Preview what fields will be written to CRM before approving — shows diff vs current CRM values ### CRM records - `search_contacts` (read) — Search contacts in the org's connected CRM (HubSpot, Salesforce, etc.) by canonical-field filters. - `get_contact` (read) — Get a single contact by its CRM-native ID. - `create_contact` (WRITE) — Create a new contact in the org's connected CRM. - `update_contact` (WRITE) — Update an existing contact's properties. - `search_companies` (read) — Search companies in the org's connected CRM by canonical-field filters. - `get_company` (read) — Get a single company by its CRM-native ID. - `create_company` (WRITE) — Create a new company in the org's connected CRM. - `update_company` (WRITE) — Update an existing company's properties. - `list_crm_platforms` (read) — List CRM platforms this server can talk to (e.g. - `get_crm_health` (read) — Check the connection health of a CRM integration for the calling org. - `list_crm_field_mappings` (read) — List the field-mapping rules between canonical field names and the CRM's native property names for a given entity type. - `list_crm_integrations` (read) — List the per-tenant CRM integrations configured for the calling org. ### Search (async) - `search` (read) — Find people or companies across the connected channels (including LinkedIn) — ONE unified verb with two modes. - `get_search_job` (read) — Poll an async job started by the `search` tool — a keyword search OR a source (people-by-role) job. - `get_content` (read) — Read posts / content for a profile or company across the connected channels (including LinkedIn). ### Entities - `list_entities` (read) — List + search canonical entities (deduplicated people / companies) in this org. - `get_entity` (read) — Get a single canonical entity by its numeric entity id (from list_entities, NOT a job id): display name, type, identifiers (email, linkedin_url, domain), most recent enrichment summary, merged source data. - `list_entity_enrichments` (read) — List every enrichment run that touched a canonical entity (by entity id from list_entities/get_entity), newest first. ### Batch enrichment - `preview_batch` (WRITE) — Preview a batch enrichment from a HubSpot list — shows entities and estimated credits - `confirm_batch` (WRITE) — Start a batch enrichment previously staged by preview_batch (by batch_id). - `list_batches` (read) — List all batch enrichment jobs for the current org, newest first. - `get_batch` (read) — Get progress for a batch enrichment by batch_id: overall status, percent_complete, plus total/completed/failed member counts. - `approve_batch` (WRITE) — Approve every completed result in a batch (by batch_id) and write them all back to the connected CRM at once. ### LinkedIn account - `get_linkedin_status` (read) — Get the org's LinkedIn connection in one read (GET /api/v2/linkedin/connection): connected/pending/disconnected/error status, profile name + URL, last verified, cookie expiry, session availability and service health. - `get_linkedin_usage` (read) — Get the org's consolidated LinkedIn usage (GET /api/v2/linkedin/usage): daily safety quotas (search/profile-view/total %), Browserbase deep-fetch budget, and the safety-deferred queue summary. - `get_profile` (read) — Composite LinkedIn profile for a person by public id — name, headline, current and past experience, education. - `get_company_profile` (read) — Company intelligence from LinkedIn by universal name (the company slug from its LinkedIn URL, e.g. - `list_linkedin` (read) — Read LinkedIn relationship + activity lists for the connected account. ### Outreach & outbound queue - `queue_outbound_batch` (WRITE) — Stage a paced batch of outbound actions (e.g. - `approve_outbound_batch` (WRITE) — Release a staged outbound batch so the dispatcher begins dripping its sends, honouring the batch schedule. - `pause_outbound_batch` (WRITE) — Pause an approved outbound batch — the dispatcher stops claiming its jobs until resumed. - `resume_outbound_batch` (WRITE) — Resume a paused outbound batch back to approved so the dispatcher continues dripping its sends. - `stop_outbound_batch` (WRITE) — Permanently stop an outbound batch: cancels every still-open job and flips the batch to stopped (allowed from staged/approved/paused). - `get_outbound_batch` (read) — Get a single outbound batch by id: the batch header (action_type, status, schedule), its jobs, and the derived per-status summary (pending / scheduled / sent / failed / cancelled). - `queue_linkedin_post` (WRITE) — Stage a LinkedIn post for review — it does NOT publish until approve_outbound_batch is called. - `update_linkedin_post` (WRITE) — Edit a STILL-STAGED LinkedIn post in place before it is approved — change any of text, visibility, content_type, media_ids, or scheduled_at. - `upload_media` (WRITE) — Upload an image/video/document to LinkedIn from a URL (the server fetches it) or a base64 blob, and get back a media_id to pass into queue_linkedin_post.media_ids (or update_linkedin_post). - `create_outreach` (WRITE) — Perform an outbound outreach / social action AS the connected account, on the channel that handles the action (today: LinkedIn). ### Account - `get_me` (read) — Get the calling user identity: Clerk user id, email, name, org membership, roles. - `get_org` (read) — Get the current organisation: id, name, slug, subscription tier, member limits.