Research Canvases

Research Canvases are project-scoped workspaces for sourcing research. Use them to capture the market map around a search before you import candidates, run checks, build shortlists, or launch outreach.

They are especially useful with Remote MCP and an external AI agent. Let the agent do the open-ended research in ChatGPT, Claude, Cursor, or another MCP-compatible client, then have it push structured findings back into TalentSourcer AI.

What a Research Canvas is for

Use a Research Canvas to keep sourcing strategy and market research in one place:

• target companies you actively want candidates from
• adjacent companies that may have relevant talent
• no-go companies you want to avoid
• role titles and excluded titles
• technologies, tools, and common synonyms
• keywords and excluded keywords
• search strings, sources, open questions, and notes

Each project can have multiple canvases. For example, you might keep separate canvases for different geographies, seniority levels, market hypotheses, or research agents.

Recommended workflow

The strongest workflow is:

1. Create or open a project in TalentSourcer AI.
2. Open Research and create a Research Canvas.
3. Connect an MCP-compatible assistant to TalentSourcer AI through Remote MCP.
4. Ask the external agent to research the market and write structured findings into the canvas.
5. Review the canvas in TalentSourcer AI, clean up anything that needs human judgment, and use the research to guide sourcing/imports.

The Research Canvas itself does not browse the web. TalentSourcer AI stores the structured output. Your external agent can do the exploratory work and use MCP tools to save the results.

Example MCP prompts

Use prompts like these with your external agent after connecting Remote MCP:

For the project "Senior Backend Engineer", create a research canvas called "Berlin fintech market map". Research likely target, adjacent, and no-go companies, then save them into the canvas with rationale, LinkedIn URLs where available, approximate headcount, and priority.

Open the research canvas for this project. Add current-or-past role titles, title exclusions, technologies, technology synonyms, keywords, and excluded keywords that would help a recruiter source strong candidates on LinkedIn.

Review this research canvas and add search strings, sources, and open questions. Keep anything uncertain as an open question instead of presenting it as verified fact.

What external agents can update

Through MCP, agents can:

• list project research canvases
• create a new research canvas
• read an existing canvas
• update canvas name or description
• add companies to target, adjacent, or no-go buckets
• update or remove companies
• add titles, technologies, synonyms, keywords, and exclusions

For technologies, save the canonical technology as technology and variants as technology_synonym with the canonical technology as the group.

Scopes

If your MCP client asks for scopes, include:

projects:read research_canvases:read research_canvases:write offline_access

Add projects:write only if the agent should create or update projects too.

In-app assistant

Inside a Research Canvas, the Research assistant can help structure notes you paste into the app. It is best for quick cleanup and data entry. For broader market research, use an external agent through MCP and let it push the structured findings into the canvas.

Best practices

• Treat the canvas as a living workspace, not a final report.
• Keep source as provenance, such as manual, assistant, or the external agent name.
• Put company websites in the company URL field and LinkedIn company pages in the LinkedIn URL field.
• Use open questions for uncertain findings.
• Review no-go companies carefully before importing or contacting candidates.