MCP: a new layer for recruiting
How recruiting teams use MCP to connect ATS, tools, and AI workflows
Opinion
Recruiting teams don’t lack tools and never did. They’re drowning in them. ATS, CRM, sourcing platforms, screening tools, interview notes, analytics tools, onboarding tools, and recruiting-related features inside HRIS systems — all with their own, often incredibly clunky interfaces and fragmented data. MCP has the potential to fix a lot of this.
MCPs could become the missing layer recruiting teams have been waiting for to operate a scattered stack of clunky UIs and make connected sense of siloed data.
What is it
MCP stands for Model Context Protocol, an open standard for AI systems such as LLMs to integrate with other tools, systems, and data sources. It was introduced by Anthropic in November 2024.
How it works
A lot of MCPs, especially those not officially provided by app developers, act as middlemen between an API and an AI tool such as Claude.
Users of Claude no longer need to write code to interact with an ATS API. Instead, they install an MCP for the respective ATS and use natural language queries such as:
“How many people applied in the last 14 days for our engineering jobs?”
The query is sent to the MCP, which has the necessary context to understand it and translate it into structured API calls, returning the answer to the user.
Why this matters for recruiting teams
Most recruiting workflows today are still UI-driven.
Example: "Get the number of applicants in the last 24 hours in Greenhouse"
- Login (session expired)
- Google SSO
- Navigate to jobs
- Find the job
- Click candidates
- Realize there’s no filter for “last 24 hours” in the UI (only via API)
- Count manually or export data
- Analyze it somewhere else
Then repeat similar steps in the next tool.
This is slow, frustrating, and prone to errors. It also makes it hard to connect the dots.
MCP introduces a different model: workflow-driven recruiting powered by AI.
Instead of navigating tools, recruiting teams orchestrate them.
Where MCPs are (and will get) powerful in recruiting
1. Pulling data from clunky systems without IT support
Many ATS platforms struggle with data visualization across pipelines, jobs, and overall performance. Even powerful products are often clunky and time-consuming to navigate.
With MCP + Claude:
- Queries like “How many people applied for job ABC last quarter vs. the previous quarter?” become trivial
- No need for API coding
- No need to wait for IT
Note: ATS vendors will likely build similar functionality over time.
2. Creating records in an ATS
Many ATS platforms lack strong bulk actions outside their APIs. MCPs can help automate bulk creation and updates via natural language workflows.
3. Connecting the dots across systems
This is where MCPs become truly powerful.
Example:
- Pull top candidates from the ATS (last 24 hours)
- Analyze resumes for patterns
- Enrich or compare with external sources (e.g. talent pools)
This might involve:
- ATS MCP
- Cloud data tools (e.g. BigQuery or similar systems)
- Additional MCPs for sourcing or enrichment
This cross-system orchestration is where traditional ATS tools will fall short.
Difference to APIs and CLIs
How is it different from an API?
An API exposes a set of endpoints.
In recruiting tech, these typically allow you to:
- Create candidates
- Retrieve jobs
- Update records
However:
- APIs are rigid and require engineering work
- Each ATS has a different API structure
That’s why middleware tools like Merge, Kombo, and StackOne exist.
MCP sits on top and makes these interactions usable via natural language.
How is it different from a CLI?
A command line interface (CLI) uses predefined commands.
Example:
list candidates
list jobs
Compared to MCP:
- CLI is more structured and limited
- Easier than APIs but still technical
- Less flexible than MCP
- Requires learning commands
MCP removes this layer entirely by allowing natural language interaction.
Things to watch
The command center
While Claude is widely used today, AI usage will likely become more verticalized.
- Teams will want department-specific AI setups
- Companies will want model independence
Relying on a single model is risky.
Abstraction layers for enterprise AI (e.g. Langdock) already reflect this trend.
Security
Many MCPs are not developed by official vendors.
This creates risk:
- Unknown data handling
- No guarantees on security practices
Enterprises must avoid using unverified MCPs for sensitive candidate data.
Garbage in, garbage out
If your tools and data are messy: MCP will not fix that.
It’s still early
MCP is still new and evolving quickly.
Early ecosystems are messy.
Think of early search engines like Yahoo or Lycos compared to Google today.
We are likely at that stage with MCP.
MCP connectors for recruiting and HR (complete list)
Below is a curated list of MCP connectors for recruiting and HR teams, including official integrations, infrastructure providers, and community-built servers.
Official and structured connectors
- Human Resources plugin by Anthropic
(Official MCP-style plugin for HR workflows inside Claude) - Slack MCP integration
(Used for internal communication workflows and automation) - Google Workspace MCP integrations
(Common for docs, sheets, and email workflows)
Recruiting and HR MCP connectors
- Greenhouse MCP via Zapier
- Workday MCP via Merge
- Workday MCP via CData
- BambooHR MCP via n8n
- Factorial MCP via Zapier
Integration-layer MCP providers (critical category)
- Merge MCP
- Finch MCP server
- Kombo integrations
- StackOne
- Knit
- CData
Unofficial and community MCP servers
Important: these are not officially verified and may carry security risks.
- Greenhouse MCP (community projects)
- Lever MCP server (community)
- LinkedIn MCP integration (community)
- Ashby MCP server (community)
- Kula MCP server (community)
There are already thousands of MCP servers in the ecosystem, but many are not audited, not secure, and not maintained.
TLDR
The MCP ecosystem is already large but highly fragmented.
The real challenge is not finding connectors. It’s choosing ones that are reliable and secure enough for production.
