Procurement is not opening fourteen rate-card tabs. They drop a compact request into an AI assistant — Claude or Microsoft Copilot: “two experienced Java developers, remote, Poland, from October, budget X”.
If the assistant can only see your website, it guesses from marketing copy. If it has an MCP connection, it asks for roles, time to first CVs, and files a structured request.
MCP is the Model Context Protocol: an open standard that lets an AI assistant call a vendor’s tool (like a function in a program) instead of only reading an HTML page. Instead of guessing from a slide that says “we do Java and SAP”, it gets fields: which role, which engagement model, how long until the first profiles.
This is not a 2028 slide. The official MCP registry already lists tens of thousands of servers. Lemon.io — a freelancer marketplace — ships “hire developers from chat”. Applicant tracking systems (ATS — the software HR uses to run hiring pipelines), including Greenhouse, Ashby and Workable, have native MCP. The channel exists. The question is whether Polish nearshore — a vendor in a nearby country, here: Poland — is visible in it, or the assistant ends on a US marketplace.
Glossary of terms in this article
- MCP (Model Context Protocol) — the protocol an AI assistant uses to call a company’s tools: roles, rates, submit a request. It is not a “Marketing Campaign Protocol”.
- T&M (Time & Materials) — hourly billing. Also called staff augmentation or body leasing: you rent a specialist for the duration of a project.
- Success fee — a commission on a successful permanent hire (ours is 15–25% of annual salary). No hire, no fee.
- Lead time — time from the request to the first profiles, or to a start date. It is not the same as “available Monday”.
- Brief — a compact request: role, seniority, start date, engagement model, company email.
- Bench — named specialists “on the books”, supposedly ready tomorrow. Commoditech does not keep one, and will not fake it in the server’s answers.
- ATS — the system HR uses to run recruitments (candidate tracking, stages, offers).
- RFP (Request for Proposal) — a formal buying document, usually from procurement.
What this channel is not
It is not a shop with people on a shelf. Commoditech does not keep a named bench of specialists ready for tomorrow morning — and we will not pretend otherwise in the server’s answers. MCP collects a qualified request: T&M or a permanent hire. We search for people on demand. First CVs are not a start date. Start follows your interviews and a contract. If a vendor returns “three senior Python, available Monday”, they are selling you CVs or lying.
1. Where the buyer actually sits
The website contact form catches marketing and sometimes HR. The staff-augmentation buyer often sits in a code editor (an IDE — integrated development environment — such as Cursor or Visual Studio Code). An engineering manager plans the quarter, the assistant sees a hole for testers (QA, Quality Assurance) or a SAP Fiori specialist (the user-interface layer in SAP), and asks “who in Poland can staff this”. If you are not in the assistant’s tools, you are not on the shortlist of vendors — regardless of your Google rank for “java staff augmentation Poland”.
The second buyer is procurement with a request for proposal. Banks and pharma companies, where we have worked for years, are starting to hand documents to assistants for a first shortlist. The assistant will not read a forty-page proposal. It calls specific fields: compliance, rates, lead time, sector references. If you do not expose those as tools, you drop out on “no data”, not on price.
2. What we expose
Public server commoditech-staffing. Assistants can:
- paste a job description and get a role match (Java is not JavaScript; a straightforward Python database app is not a document-search project, so-called RAG — Retrieval-Augmented Generation),
- get a range for time-to-first-CVs and time-to-start — separately for staff augmentation and for a permanent hire,
- get hourly bands or the success-fee policy (15–25%, no hire no fee),
- submit a request to a company email; a human replies within 24 business hours.
Candidate data does not leave. The request does not write straight into the CRM (the system where we keep client relationships). Full CVs still go by email, after consent, as they have since 2012. How to connect and which tools are available: MCP for AI agents.
Hourly bands are the T&M rate card, not tomorrow’s invoice. The market map — what others post on No Fluff Jobs and at Hays — is separate: IT body leasing rates in Poland 2026.
3. Staff augmentation and permanent hire in one channel
Historically about 98% of our opportunities are Time & Materials — renting specialists by the hour. Permanent recruitment is coming back, because firms freeze junior roles and open them for SAP, architecture and GRC (Governance, Risk, Compliance — corporate controls, risk and regulatory fit). An assistant that gets “hire an experienced Fiori specialist on payroll” must not receive an hourly card. It gets a success-fee model and a longer lead time (notice periods).
Same search engine: contractor network plus direct search. Different contract, different SLA (Service Level Agreement — the promised service level), different cash. MCP distinguishes time_and_materials from recruitment. If the request is ambiguous, a human asks — the tool does not guess.
The demand split (frontend drowning in CVs, SAP sitting empty) does not vanish because an assistant filed a request. It stays. What changes is the request: role, seniority, start date, engagement model, company email. Better than a “write a message” form.
4. What is sales, what is marketing
Volume will be small for the first months. Tire-kickers, competitors, one real request. That is fine. The server is cheap to run. The value is triple:
- Sales funnel — a qualified request with no salesperson at the door. Success of the first version: one such request that became a call.
- Demand signal — even without a submitted brief. If forty assistants a week ask for SAP consultant rates, you start searching SAP people before the formal RFP arrives.
- AI-service demo — “we will build you the same kind of server, so your customers’ assistants can query your catalogue”. That is T&M or a fixed price on engineers we already want to sell. A live server beats a slide.
We do not charge for tool calls. An API key (an access password for the programming interface) on day one would kill discoverability. A paid market report — only after the logs show volume, not before.
5. How not to break it
- Do not ship a fictional bench to production. Trust in the assistant channel dies on the first invented number of people.
- Do not auto-quote a contract. Ranges and a “this is not a quote” note in every tool response.
- Do not mix the internal CRM with the public server. Contact search on the internet is an incident, not a feature.
- Do not assume assistants will find you. There are about 100 thousand servers in the indexes. You need a listing, a llms.txt file (a short instruction on the website for language models), this article, and clear calls to action on the staff-augmentation landing pages.
Connect and ask — or have us build yours
Staff augmentation or a permanent hire: drop a job description into an assistant with our MCP, or write to us. An equivalent server for your vendor catalogue, knowledge base or buying process: same team, T&M or a fixed price. We are not selling a slide about agents. We are selling a server that already answers.