Ashby
by Ashby
Query your Ashby applicant-tracking system from an AI assistant — candidate profiles, pipelines, interview briefings, offers and your own scorecards — and take four documented write actions on candidate records. Eighteen documented tools, four of which write. Every read is scoped to the signed-in user's own Ashby permissions.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect Ashby via MCP
https://mcp.ashbyhq.com/mcp/v1Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Ashby Tools & Capabilities (16)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- Four documented tools write, and Ashby says more are coming.
create_candidate,consider_candidate_for_job,change_application_stageandadd_note_to_candidatechange records. Ashby's documentation states it adds tools over time including additional write actions. Read-shaped names on the other twelve should not be generalised across the surface. - The grant cannot be narrowed. One
mcpscope covers reads and writes over candidate data. There is no read-only connection available to a client that wants one. - The directory record carries no permissions label at all. There is no field to falsify and none to rely on. Read Ashby's tool list instead — it is the accurate source, and it distinguishes the writes.
- The directory listing is one tool short and one action long.
submit_mcp_feedbackis documented but unlisted; the directory description mentions sending emails, which no documented tool does. - We could not verify the tool list or its behaviour ourselves. The endpoint returns HTTP 401 to an anonymous handshake, so no schemas and no safety annotations were readable. Every classification here is Ashby's documentation, restated.
- The permission claim is Ashby's, not our observation. We did not authenticate, so we could not test what a tool returns when asked for a candidate the user cannot see. Ashby does not document that behaviour either, nor a field-level map from tools onto permission checks.
- Reach scales with the connecting user's role, not with the connector. An Org Admin holding the global confidential-roles permission reaches every confidential job; a Limited Access user reaches almost nothing. The same server behaves very differently on two accounts.
- Offers are reachable, and compensation rides on the objects around them.
filter_recordsandget_record_detailsboth support offers. Ashby stores compensation on jobs and job postings, which are also reachable. Plan the allow-list accordingly. - The server is beta and the contract is explicitly unstable. Ashby warns that MCP tool inputs and outputs may change without notice and points anyone needing a stable contract at its public API.
- Rate limits are real and published. 120 requests per minute per auth token, and 120 tool-budget units per minute per user-and-organization pair, with most tools costing 1 unit and
filter_recordscosting 2. An agent fanning out parallel filters will hit the second limit before the first. describe_object_fieldsis not meant to be called directly. Ashby describes it as a behind-the-scenes helper thatfilter_recordsdepends on. It is listed as a tool, so a client will surface it, but its output is schema discovery rather than an answer.- The subjects are third parties. Candidate records concern people who applied for a job and are not parties to the connector. Nothing on this page is legal advice, but the data class differs from a CRM's and an organisation's own policy should decide what an agent may do with it.
Frequently asked questions
Whose interview feedback can the Ashby MCP connector read?
Two different answers, depending on the tool. get_submitted_feedback returns only scorecard feedback you personally submitted — Ashby states plainly that it does not show other interviewers' feedback. But get_candidate pulls interview feedback from a candidate's profile generally, bounded by your Ashby permission level rather than by authorship. A Hiring Team Member sees the panel's feedback there.
Does the Ashby MCP server write to your ATS?
Yes. Four of the eighteen documented tools write: create_candidate makes a new candidate record, consider_candidate_for_job creates an application, change_application_stage moves a candidate through the hiring process, and add_note_to_candidate drafts a profile note. Ashby documents the note tool as requiring your confirmation before saving, and says more write actions are planned.
Can the Ashby MCP server see candidates I cannot see in Ashby?
No, according to Ashby. Each user completes an individual OAuth flow, and Ashby states the server only returns data that user can already see. A user not involved in a hiring process cannot query those candidates or roles. We could not test this ourselves — the endpoint is OAuth-gated and we do not authenticate — so the claim is Ashby's, restated.
Can the Ashby MCP connector reach offers and compensation?
Offers, yes. Ashby documents filter_records and get_record_details as supporting offers alongside candidates, applications, jobs, job postings, openings and scheduled interviews. Compensation is not named as a separate record type, but Ashby stores compensation on jobs, postings and offers, so amounts can surface through those objects within your existing permissions.
How many tools does the Ashby MCP server actually have?
Ashby's own documentation describes eighteen; Anthropic's directory snapshot of 2026-08-16 lists sixteen. Seventeen of Ashby's eighteen are named tools, and the one the directory omits is submit_mcp_feedback, which sends notes to Ashby's team. Ashby also states it adds tools over time, including further write actions, so any fixed count is a snapshot rather than a ceiling.
What Ashby plan and role do you need for the MCP server?
All Ashby plans include it — Foundations, Legacy Plus, Plus and Enterprise — but Analytics-only organizations are excluded. An Org Admin must first switch on the MCP Server toggle under Admin, Organization Setup, Opt-In Features. After that, Ashby documents Elevated Access users as the ones who can connect their own account.
Is the Ashby MCP server stable enough to build on?
Ashby's own documentation says no. The page is titled a beta, and Ashby warns that MCP tool inputs and outputs may change without notice, directing anyone needing a stable contract to its public API instead. Tool shapes can therefore move under a connected agent, and Ashby publishes no deprecation window for the MCP surface.
Sources
- Ashby MCP Server (Beta) — Ashby's MCP documentation, reached via the
https://docs.ashbyhq.com/mcpredirect. Fetched 2026-08-19. · retrieved 2026-08-19 - User Permissions — Ashby's global roles and access roles. Fetched 2026-08-19. · retrieved 2026-08-19
- What is a confidential job? — confidential role visibility and the Org Admin override. Fetched 2026-08-19. · retrieved 2026-08-19
- Compensation — where Ashby stores compensation data. Fetched 2026-08-19. · retrieved 2026-08-19
- Anthropic MCP directory listing for Ashby — tier, tool names, categories. Snapshot dated 2026-08-16. · retrieved 2026-08-16
- Live protocol checks against
https://mcp.ashbyhq.com/mcp/v1, its RFC 9728 descriptors in three path forms, and the RFC 8414 and OpenID Connect metadata athttps://mcp-auth.ashbyhq.com/oidc. Performed 2026-08-19. No authentication was attempted and no tool was called. · retrieved 2026-08-19
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Productivity
- Developer
- Ashby
- Tools
- 16
- Domain
- mcp.ashbyhq.com
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.