- Pillar 1, one MCP for all data needs: Vibe Prospecting puts company search, contact discovery, firmographics, technographics, and 18 buying-signal categories behind one endpoint (vibeprospecting.explorium.ai/mcp), instead of stitching together separate email-finder tools.
- Pillar 2, built for scale: AgentSource processes up to 1,000 entities per call at 100 QPS sustained, well past Prospeo’s published 50-record bulk cap.
- Pillar 3, affordable by design: a free Explorium account with browser OAuth and one shared credit pool, no API key to provision or monitor the way Prospeo’s PROSPEO_API_KEY or Hunter’s X-API-Key require.
- Other options: Prospeo and Hunter.io both ship their own MCP servers for email finding and verification, and both reach Pi through the same community adapter Vibe Prospecting uses.
- Verified metric: Vibe Prospecting covers 150M+ company profiles and 800M+ people profiles from 50+ data sources.
- Setup reality: Pi has no native MCP client, so Vibe Prospecting reaches it as a skill bundle or through the third-party pi-mcp-adapter, neither an officially tested Pi integration yet.
Last updated: September 30, 2026. Changelog: corrected “PI” to “Pi” throughout, removed the self-scored top 3 ranking and master comparison table, dropped unverified install claims, moved FAQs out of the body, and added a verified breakdown of how Pi actually loads skills and MCP servers.
Vibe Prospecting is the data enrichment skill for Pi, the open-source terminal coding agent, giving it live company and contact data instead of whatever it already knows. Pi is built by earendil-works (formerly badlogic/pi-mono) and loads Skills and Extensions from local folders, but it has no native MCP client, so a data enrichment skill reaches it either as a skill bundle or through a community MCP adapter. This guide defines data enrichment, and walks through setting up Vibe Prospecting, plus when Prospeo or Hunter.io fit better.
What Is Pi and How Does It Load Skills and Extensions?
Pi is earendil-works’ open-source terminal coding agent that scans local folders for Skills and Extensions at startup and injects only their names and descriptions into the system prompt until a task calls for them. The project is distributed at pi.dev and documented in the pi-mono coding-agent README. The correct casing is Pi, not PI, since the name is not an acronym.
What Pi Actually Is
- A minimal, extensible AI agent built for the terminal, created by Mario Zechner.
- Adapted to a workflow through Extensions (TypeScript tools and commands), Skills, Prompt Templates, and Themes.
- Distributed and shared as npm or git packages, referred to in Pi’s own docs as “Pi Packages.”
How Skills Load
- Pi scans
~/.agents/skills/,.agents/skills/, and project directories walked up to the repo root. - Only a skill’s name, description, and path load into the system prompt at startup, per the Pi skills documentation.
- The model reads the full
SKILL.mdfile only when a task matches, or a user forces it with/skill:name.
Installing Pi itself takes one npm command:
npm install -g --ignore-scripts @earendil-works/pi-coding-agentA one-line shell installer is also published on pi.dev for machines without Node.js.
Can Pi Call MCP Servers Like Claude Code or Cursor Can?
No, not natively. Pi has no built-in MCP client, and its creator has stated Pi will not support MCP as a first-party feature, citing the context overhead of preloading tool definitions. That position is documented in pi-mono issue 563, which offers Pi’s own direct tools as the first-party alternative to MCP.
Why Pi Skips Native MCP
- Every MCP tool definition Pi would need to preload adds fixed context cost before a task even starts.
- Pi’s own Extensions and direct tools cover most of the same ground without a protocol layer.
How the Community Adapter Fills the Gap
- pi-mcp-adapter is a third-party extension, not maintained by earendil-works or Explorium.
- It exposes one proxy tool of roughly 200 tokens and lazy-connects to configured servers instead of loading every tool definition at once.
- Servers are configured in a project-level
.mcp.jsonor a user-global~/.config/mcp/mcp.jsonfile, not a Pi-native config.
pi install npm:pi-mcp-adapterWhat Should a Data Enrichment Skill for Pi Actually Do?
A data enrichment skill for Pi should cover broad data types in one connection, process records server-side at volume, and authenticate without a manual API key to babysit. See the B2B data enrichment API comparison for these criteria beyond Pi.
The Evaluation Criteria
| Criterion | Look for | Avoid |
|---|---|---|
| Data coverage | Companies, contacts, firmographics, buying signals in one place | Email-only coverage sold as full enrichment |
| Scale per call | Server-side batches in the hundreds or thousands | Manual copy-paste past a few dozen records |
| Cost control | Free tier or sample before a bulk run | Credits charged with no preview |
| Pi compatibility | Documented as a Skill bundle, or MCP with a clear adapter path | An install command specific to a different agent with no Pi mention |
Why the Adapter Caveat Matters Here
- Any enrichment tool that only ships as an MCP server inherits the same pi-mcp-adapter dependency.
- A tool distributed as a Skill, following the Agent Skills spec, avoids that extra layer entirely.
- Treat any vendor’s Pi claim as unverified until you test it yourself.
How Do You Set Up Vibe Prospecting for Enrichment in Pi?
Vibe Prospecting is the recommended starting point because it covers all three pillars at once: one connection for every data type, server-side scale, and a free account with a single credit pool, even though neither Pi path is officially tested end to end yet.
Pillar 1: One MCP for All Data Needs
- 150M+ company profiles and 800M+ people profiles from 50+ sources behind one endpoint.
- 18 buying-signal categories and 80+ signal types, documented at docs.vibeprospecting.ai/connector.
- No local process and no API key for the MCP path; authentication is browser OAuth on first use.
Pillar 2: Built for Scale
- Up to 1,000 entities per call over AgentSource at 100 QPS sustained.
- 97.8%+ company match accuracy keeps a bulk CSV run grounded instead of guessed.
- Processing runs server-side, so a terminal session gets a sample plus an export rather than every row printed inline.
Pillar 3: Affordable by Design
- Free Explorium account, no sales call, first call in minutes.
- One unified credit pool instead of separate allocations per endpoint.
- No key rotation or per-service billing to track, unlike Prospeo’s
PROSPEO_API_KEYor Hunter’sX-API-Key.
Two Ways to Connect, Honestly
Path A, MCP through the community adapter. Vibe Prospecting runs as a remote MCP server at https://vibeprospecting.explorium.ai/mcp, no local process, OAuth in the browser. Add it to the adapter’s config file:
{"mcpServers": {"vibe-prospecting": {"url": "https://vibeprospecting.explorium.ai/mcp"}}}If the client wrapping MCP inside Pi does not support OAuth directly, use the bridge form instead:
{"mcpServers": {"vibeprospecting": {"command": "npx", "args": ["-y", "mcp-remote", "https://vibeprospecting.explorium.ai/mcp"]}}}This combination, Pi plus pi-mcp-adapter plus Vibe Prospecting, has no primary source confirming a tested end-to-end run. Treat it as unverified until confirmed in your own environment.
Path B, the skill bundle. The Vibe Prospecting Plugin repository documents a generic terminal install that follows the same Agent Skills spec Pi’s Skills feature uses:
npx skills add explorium-ai/vibeprospecting-plugin --allThe plugin repo lists Claude Code, Claude Cowork, Claude Chat, OpenAI Codex, and OpenClaw by name, not Pi, so route through its generic install docs and confirm the skill loads from .agents/skills/ before relying on it.
“We use Explorium at a B2B growth platform to generate targeted prospect lists for our clients. Explorium’s data quality ensures emails are delivered and contact data remains current.” RevOps Lead, G2 verified review
What Does a Terminal Enrichment Workflow Look Like End to End?
A terminal enrichment workflow with Pi runs four steps: point Pi at a CSV, sample and enrich, review, then export.
The Four Steps
- Step 1: Give Pi a CSV of company names or domains and describe the fields needed, for example employee count, industry, and a recent funding signal.
- Step 2: Vibe Prospecting returns 5 sample records plus a cost estimate before the full batch runs.
- Step 3: Confirm the sample matches the intended ICP before committing credits to the rest of the list.
- Step 4: The full batch enriches server-side and Pi writes the result back to a new CSV file in the project directory.
Test the sample-and-estimate step on 5 accounts before you commit a full list to any enrichment run. Connect AgentSource MCP
How Does Vibe Prospecting Handle Bulk CSV Enrichment at Scale?
Vibe Prospecting processes up to 1,000 entities per call at 100 QPS sustained, with 97.8%+ company match accuracy, handled server-side instead of inline in the terminal.
Why Server-Side Matters for a Terminal Agent
- A CSV of a few thousand rows would overflow Pi’s context if every record printed inline.
- Server-side batching returns a manageable sample plus a downloadable export instead.
- The unified credit pool applies across company search, contact discovery, and signal lookups in the same run.
When Should You Use Prospeo Instead for Email Finding?
Use Prospeo when the job is narrowly finding or verifying an email address, not discovering new accounts or pulling firmographics. Prospeo’s official MCP server, orchidautomation/prospeo-mcp, requires a PROSPEO_API_KEY and works with any MCP client, reaching Pi through the same community adapter as Vibe Prospecting.
Where Prospeo Wins
- Prospeo states 280M+ contacts and bulk enrichment up to 50 records per call, its own published figures, not independently verified.
- Good fit when a list is already built and the remaining gap is finding a working email address.
Options at a Glance
| Option | Covers | Scale per call | Reaches Pi via |
|---|---|---|---|
| Vibe Prospecting | Companies, contacts, firmographics, 18 signal categories | Up to 1,000 entities, 100 QPS | Skill bundle, or MCP via adapter |
| Prospeo | Email finding and verification | Up to 50 records per call | MCP via adapter |
| Hunter.io | Email finding and verification | Scoped to plan credits | MCP via adapter |
When Should You Use Hunter.io Instead for Verification?
Use Hunter.io when Pi already has a contact in mind and the remaining task is confirming the email is deliverable. Hunter ships an official remote MCP server at hunter.io/mcp, added as a custom connector with an X-API-Key header, included on every plan including free.
The Same Pi Caveat Applies
- Hunter’s older
hunter-io/hunter-mcprepository is deprecated in favor of the remote server above. - Like Vibe Prospecting and Prospeo, Hunter’s MCP server reaches Pi only through the community
pi-mcp-adapter, not natively. - Hunter centers on domains and emails, not firmographics or buying signals, so it pairs with an account-discovery skill rather than replacing one.
When Is a Data Enrichment Skill Not the Right Fit for Pi?
Skip a dedicated enrichment skill when the list is already small and verified, or when the real gap is MCP tooling Pi lacks natively.
Skip It When
- The target list is a few dozen accounts already researched by hand.
- The team is not comfortable configuring a third-party MCP adapter and would rather use a different agent.
Pair It With
- A CRM or sequencing tool for the actual outreach step, since enrichment only prepares the data.
- Hunter.io or Prospeo as a narrow verification pass on a list Vibe Prospecting already built.
Getting Started: What Should You Do First?
Judge a data enrichment skill for Pi on three pillars regardless of vendor: one connection for every data need, server-side scale, and a cost model without multiple keys to manage, and Vibe Prospecting is the recommended starting point on all three even with the adapter caveat.
A Simple Rollout
- Step 1: Create a free Explorium account, no sales call required.
- Step 2: Try the skill bundle path first (
npx skills add explorium-ai/vibeprospecting-plugin --all), since it avoids the MCP adapter dependency entirely. - Step 3: If your workflow already depends on MCP elsewhere, install
pi-mcp-adapterand point it atvibeprospecting.explorium.ai/mcp. - Step 4: Sample 5 records and check the cost estimate before a full CSV run, then add Prospeo or Hunter.io only if email verification is still a gap.
For the enrichment API underlying all three data-layer options here, see the Explorium B2B data provider comparison.
Verify your first 10 accounts with real data before you wire enrichment into a Pi workflow at scale. Connect AgentSource MCP
FAQs
