OnFire MCP Plugin
The OnFire MCP Plugin is a Claude Code plugin that ships a curated set of OnFire skills. A skill is a high-level recipe — orchestration, rendering contract, and behavioural rules — built on top of one or more OnFire MCP tools. The MCP server owns the data; the plugin owns how an agent should use that data well.
What you get
Section titled “What you get”The plugin contributes ten skills, grouped by what the user is trying to do:
| Skill | What it does |
|---|---|
onfire-prospecting | Orchestrator for the five atomic prospecting skills. Routes work end-to-end. |
ai-prospecting | Atomic: run AI prospecting or score one named person. |
match-company | Atomic: resolve a company to a verified LinkedIn identity. |
match-person | Atomic: resolve a person to a verified LinkedIn profile. |
contact-data-enrichment | Atomic: append verified emails / phones. Enforces the two-phase consent flow. |
manage-ai-prospecting | Atomic: shadow lifecycle for the prospecting schema (draft, test, ship, discard). |
bdr-account-research | Generate the customer-facing BDR account research HTML report. |
account-expansion-plan | Generate a single-account Expansion Opportunities Plan (CRM + prospecting + signals). |
poc-intelligence | Generate a tenant-wide POC Intelligence Report (funnel + plays + playbook). |
prospect-signals-bot | A Slack-channel watcher that posts a minimal new-business signal summary per request. |
Why a plugin and not just an MCP server?
Section titled “Why a plugin and not just an MCP server?”The MCP server is stateless and policy-light by design — it exposes typed tools, scopes them to the authenticated tenant, and returns structured JSON. That’s the right shape for any agent runtime to consume.
The plugin layers on top:
- Orchestration. Which tool to call first when you only have a name. When to poll vs. retry. Which dataset to hand off to which next tool.
- Rendering contracts. A BDR brief / expansion plan / POC report are HTML deliverables with strict layout and branding rules. The plugin enforces those rules so the agent produces a consistent artifact each time.
- Behavioural rules. Required user-approval phrases for paid tools.
How to surface the
download_urlfromdownload_dataset. How to re-format reasoning sentinels for the user.
In short: the MCP server is for any agent; the plugin is for agents that need to produce OnFire-shaped deliverables without rediscovering the rules every time.
Which tools each skill leans on
Section titled “Which tools each skill leans on”| Skill | Primary MCP tools |
|---|---|
onfire-prospecting | match_company, match_person, ai_prospecting, contact_data_enrichment, manage_ai_prospecting, query_datasets, describe_dataset, download_dataset, list_datasets |
ai-prospecting | ai_prospecting, ai_prospecting_field_glossary, dataset companions |
match-company | match_company |
match-person | match_person |
contact-data-enrichment | contact_data_enrichment |
manage-ai-prospecting | manage_ai_prospecting, ai_prospecting (shadow runs) |
bdr-account-research | bdr_account_research, query_datasets, the typed follow-up tools |
account-expansion-plan | get_tenant_settings, query_intent_signals, ai_prospecting — plus the CRM integration proxy outside this tool list |
poc-intelligence | get_tenant_settings, list_datasets, query_datasets, describe_dataset, search_people, ai_prospecting — plus the CRM integration proxy, Metabase, and Snowflake outside this tool list |
prospect-signals-bot | query_intent_signals — plus Slack outside the MCP |
Install
Section titled “Install”The plugin is distributed as a .plugin file. Drop it into Claude Code
the same way you install any other plugin and the skills become
available immediately. The plugin assumes you’ve already connected the
OnFire MCP server — there are no extra credentials to manage.