Databar (MCP) Skill

Run Databar data workflows through the mcp__databar__* MCP tools: discover and run enrichments and waterfalls, manage workspace tables, build reusable flows, and export results to CRMs. All paid operations bill in Databar credits (not dollars) — this skill prices before it spends.

When to Use

• Contact enrichment and lead generation: find/verify emails, phone numbers, LinkedIn profiles, decision-makers at companies.
• Web data pulls: Google Maps, SERP, reviews, job postings, Amazon, YouTube search/comments.
• Company intelligence: tech stack, firmographics, hiring signals, funding/partnership news by domain.
• Marketing & SEO: competitor keywords (organic + PPC), transactional keywords by topic, LinkedIn ad library, brand social stats.
• Social & community intelligence: X/Twitter, TikTok, LinkedIn posts, Reddit, Telegram, Hacker News.
• Job search & recruiting: people search by title/company, hiring-company discovery, Sales Navigator exports.
• E-commerce & reputation: product details/pricing from any store, Google Shopping, Trustpilot/Capterra/app-store reviews.
• Investor & market intelligence: funding rounds and investors, public-company financials, earnings transcripts, real-time stock data.
• App-market intelligence: App Store/Google Play data, revenue estimates, review mining.
• Web traffic & domain intel: traffic estimates, SEO statistics, global rank, redirects, cheap firmographics by domain.
• Repeatable pipelines: tables with attached enrichment columns, scheduled sources, flows.
• Pushing results to HubSpot, Salesforce, Google Sheets, webhooks.

Don't use for: one-off page reads that web_extract already covers, or any bulk run where you haven't checked price_credits yet.

Prerequisites

• Databar MCP server connected. Verify: terminal(command="hermes mcp list") shows databar ... enabled.
• Hosted endpoint: https://mcp.databar.ai/mcp (preferred — always current). API key from your Databar workspace → Integrations, sent as Authorization: Bearer <key>.
• Credits in the account. Check before spending: mcp__databar__get_user_balance.

How to Run

Databar tools are loaded on demand. In Hermes: tool_search(query="databar <topic>") → tool_describe(name="mcp__databar__<tool>") → tool_call. Tool names are always mcp__databar__<tool>.

The Databar Model (read once)

1. Discover → run → poll → persist (optional) → export (optional).
2. An enrichment is one provider. A waterfall tries multiple providers in sequence until one returns data — use waterfalls for email/person lookups where hit rate matters.
3. Every paid call has a price_credits cost (visible in get_enrichment_details), billed per record. Paginated enrichments bill per page.
4. Runs are async: run_* returns a task id, not data. get_task_status returns results once complete.
5. Tables: create → add rows → attach enrichment/waterfall/flow as result columns → run on all or empty rows.

Quick Reference

Procedure — one-off enrichment

1. get_user_balance — record starting credits. ✓ balance noted.
2. search_enrichments(query="...") — shortlist by rank; note id, price_credits, auth_method per option.
3. get_enrichment_details(enrichment_id=...) — exact param slugs and response fields. If a param shows choices.mode: "remote", resolve options with get_param_choices. ✓ params known before spending.
4. run_enrichment(enrichment_id, params) or run_bulk_enrichment(enrichment_id, params_list) for batches.
5. get_task_status(task_id) — wait ~3–5 s for singles, then poll every few seconds; waterfalls take longer.
6. Bulk results are position-aligned: data[i] answers params_list[i], null means no data. ✓ len(data) == len(params_list).

Procedure — email/contact waterfall

1. search_waterfalls(query="email finder") — note identifier, available_enrichments, input_params.
2. run_waterfall(waterfall_identifier="email_getter", params={...}) — optionally pass email_verifier (an enrichment id) to verify hits, or provider_ids to restrict providers.
3. Poll get_task_status; report hit rate. For batches use run_bulk_waterfall (same position-alignment rule).

Procedure — table pipeline (repeatable)

1. create_table(name, columns) then create_rows (max 100/call; options.allow_new_columns: true auto-creates unknown columns).
2. add_table_enrichment(table_uuid, enrichment_id, params) — each param maps to {"type": "mapping", "value": "<column name>"} (read per-row) or {"type": "simple", "value": "<static>"}.
3. run_table_enrichment(table_uuid, enrichment_id, run_strategy="run_empty") — skips rows that already have results; "run_all" re-processes everything (billed again).
4. Read back with get_table_rows (filters: equals, contains, not_equals, is_empty, is_not_empty — AND-ed; max 500/page).
5. Export: search_exporters → get_exporter_details → add_table_exporter → run_table_exporter.

Recipes for common goals are in references/ — recipes.md (sales/lead-gen core: lead lists, Maps, hiring signals, tech stacks, scheduled sources), recipes-marketing-social.md (SEO/PPC, ads, social listening, review mining, price watch, news), recipes-people-research.md (decision-makers, warm intros, job search, dossiers, research sourcing), and recipes-finance-app.md (funding intel, public-company teardowns, app-market intelligence, domain/traffic teardowns).

Cost discipline

• Call get_enrichment_details before the first run of any enrichment — the same data can cost 1–15+ credits depending on provider. Price the whole chain, not one hop.
• Paginated enrichments bill per page — request only the pages you need.
• run_strategy="run_empty" instead of "run_all" on re-runs of table enrichments.
• Report the credit delta (get_user_balance before/after) with the delivered result.

Pitfalls

• Async ≠ done. A task id from run_* is not a result. Always confirm via get_task_status before reporting data.
• Positional joins. Bulk results carry no keys — preserve your input order or join by index.
• Destructive tools are permanent. delete_table, delete_rows, delete_column, delete_flow, delete_folder have no undo. Never delete without explicit user instruction.
• delete_flow is refused while the flow is attached to a table column — remove the column first.
• upsert_rows key must be exactly one column; create_rows/upsert_rows/patch_rows cap at 100 rows per call.
• BYOK enrichments (auth_method: "user_or_databar") may require your own provider API key if Databar-side auth fails.
• Results may come from cache; pass skip_cache: true when freshness matters (full price).
• The npm databar-mcp-server stdio package lags the hosted server — prefer the hosted URL unless your client requires stdio.

Verification

• get_user_balance before/after shows the expected credit delta.
• get_task_status reports completed with data; for tables, get_table_rows shows result columns filled on the intended rows.