TalentSourcer AITalentSourcer AI
Agentic Use

Remote MCP

Connect MCP-compatible AI assistants to TalentSourcer AI.

Use the remote MCP endpoint when your AI client supports adding an MCP server by URL.

https://api.talentsourcer.ai/mcp

When your MCP client connects, TalentSourcer AI uses OAuth with PKCE. You will be redirected to the TalentSourcer AI app, where you can:

  1. sign in
  2. choose the workspace to connect
  3. review requested scopes
  4. approve or deny access

After approval, the MCP client receives an OAuth access token. The token is scoped to your user, the selected workspace, and the requested permissions.

Device flow

Some agents connect MCP servers through OAuth device flow instead of an automatic browser redirect. For those agents, send a prompt like this:

Connect the TalentSourcer AI MCP at https://api.talentsourcer.ai/mcp. It supports OAuth device flow. Discovery is available at https://api.talentsourcer.ai/.well-known/oauth-protected-resource. Run the device flow and send me the authorization link like: Authorize here: https://next.talentsourcer.ai/device?code=XXXX-XXXX. Then poll the token endpoint until authorization completes.

When the agent replies with an authorization link, open it, sign in, choose the workspace, and approve access.

Metadata endpoints

Most MCP clients discover these automatically:

https://api.talentsourcer.ai/.well-known/oauth-protected-resource
https://api.talentsourcer.ai/.well-known/oauth-authorization-server

Device-flow capable clients can also use:

https://api.talentsourcer.ai/oauth/device_authorization
https://next.talentsourcer.ai/device

MCP tools

The remote MCP server currently exposes tools for:

  • projects
  • project-scoped Research Canvases for sourcing research
  • search creation, updates, and candidate import
  • search, shortlist, and campaign candidate listing/removal workflows
  • candidate private context updates
  • briefings and requirement updates
  • talent pool insights
  • candidate checks
  • shortlists
  • draft campaigns and campaign analytics
  • outreach task review and manual task state updates

If your MCP client asks for scopes, use the smallest set required for your workflow. For broad recruiting-agent workflows, request:

projects:read projects:write briefings:read briefings:write research_canvases:read research_canvases:write candidates:read candidates:write shortlists:read shortlists:write campaigns:read campaigns:write outreach_tasks:read outreach_tasks:write inbox:read ats:read ats:write offline_access

For research-heavy workflows, Research Canvases are usually the best place for an external agent to write its findings. Ask the agent to create or update a project canvas with target, adjacent, and no-go companies, role titles, technologies, search strings, sources, and open questions. Include these scopes:

projects:read research_canvases:read research_canvases:write offline_access

Candidate private context

When importing candidates from an ATS or another private source, agents can add concise organization-scoped context through add_candidates_to_search or update it later with update_candidate_context.

  • Use notes for short recruiter/ATS observations, ideally 3-8 role-relevant bullets such as call notes, motivation, availability, compensation expectations, or dealbreakers.
  • Use resumeData for a concise factual CV/resume/ATS profile summary, such as relevant work history, projects, skills, education, and certifications.
  • Do not paste raw ATS exports, full transcripts, email threads, or unrelated personal data. Summarize relevant evidence.
  • These tools do not start candidate checks. Private context may be used only for BYOK candidate checks after a user starts checks in the TalentSourcer UI.

Troubleshooting

If MCP authorization opens the wrong workspace, use the workspace selector on the authorization page before approving.

If a command returns a permission error, reconnect with the missing scope.

On this page