volo-mcp

Give your agent VOLO's evidence base on how AI, robotics and automation are changing specific jobs — task by task, with sources.

VOLO publishes, for 70 occupations and 14 university majors, what is changing in a specific job, what the evidence for that is, and what the evidence does not establish. Every judgement is labelled evidenced or inferred; every verified record carries its source, scope and dates. This package exposes all of it to any MCP client.

v2.0.0 is a different product from v1.x. The 1.x releases were an aviation booking server for an earlier VOLO; that product was retired in September 2026 and its endpoints answer 410. If you installed 1.x, upgrade — the tool names, schemas and purpose have all changed.

Two ways to connect

Remote — any MCP client that speaks Streamable HTTP (recommended):

https://mcp.flyvolo.ai/mcp

Public, read-only, no key, JSON-RPC 2.0, protocol 2025-06-18. Registry name ai.flyvolo/knowledge.

Local — clients that only speak stdio (Claude Desktop, Cursor, Claude Code, CLI tools):

npx volo-mcp

This package is a thin bridge: it forwards every request to the remote server and holds no logic of its own, so it can never drift from what flyvolo.ai shows. Override the endpoint with VOLO_MCP_URL if you run a mirror.

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) · %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "volo": { "command": "npx", "args": ["-y", "volo-mcp"] }
  }
}

Cursor · Claude Code · other stdio clients

Same shape: command npx, args ["-y", "volo-mcp"].

Claude API (MCP connector — no local install)

{
  "mcp_servers": [{ "type": "url", "url": "https://mcp.flyvolo.ai/mcp", "name": "volo" }],
  "tools": [{ "type": "mcp_toolset", "mcp_server_name": "volo" }]
}

with beta header anthropic-beta: mcp-client-2025-11-20. No authorization_token is needed.

OpenAI Responses API

{ "type": "mcp", "server_label": "volo", "server_url": "https://mcp.flyvolo.ai/mcp", "require_approval": "never" }

Tools

Sixteen, all read-only. locale accepts en or zh (the site itself is in en · zh · ja · es · pt; tool output in the other three is on the roadmap). Start with resolve_occupation_or_major whenever a person names their job or degree.

Tool	Arguments	What it returns
resolve_occupation_or_major	query, locale?	Resolve a job title, a company's internal name for a role, a nickname or a degree name to a canonical VOLO entity.
search	query, locale?, limit?	Search occupations, majors and verified change records by keyword.
get_occupation	slug, locale?	Full task-level assessment of one occupation.
get_role_tasks	slug, locale?	The task breakdown for one occupation.
find_evidence	slug, taskId?, locale?	Verified evidence records for an occupation, optionally for one task.
get_change	id, locale?	One verified change record by id, with VOLO's reading: what it means, what it does not yet establish, and what a reader can verify themselves.
find_affected	id, locale?	Which occupation and which specific tasks a change record bears on.
get_change_history	slug, locale?	The verified change timeline for an occupation, oldest first.
find_transition_paths	slug, locale?	Options for someone in this occupation.
list_business_functions	locale?	For the business side: the functions a company is organised into, and which occupations sit in each.
get_major	slug, locale?	A degree and the occupations it feeds.
find_majors_for_occupation	slug, locale?	Which degrees lead into this occupation.
find_notes	query?, locale?	VOLO's long-form answers to the big questions — how many jobs AI has actually taken, what actually gets automated, how long someone has, what AI law already changed, where AI lands in a company.
get_note	slug, locale?	The full text of one note, section by section, each section with the verified records it rests on.
find_notes_for_occupation	slug, locale?	Which notes argue from this occupation's own records or task judgements.
describe_coverage	locale?	What VOLO covers and what it does not, including which occupations have no evidence yet.

Every task comes back with its basis and its limits; every record with its scope, stage and source tier. Carry those into the answer — an inferred judgement presented as a finding misrepresents the source.

What these tools will not return

• A number of years. "How long do I have?" is answered with a signal the person can check, never a date.
• A risk score, probability of job loss, or readiness rating for a person, occupation or major.
• A recommendation about who to make redundant, or which vendor to buy.
• A verdict on whether a forecast came true. Forecasts are recorded with who said them and when; the arithmetic is left to the reader.

These are enforced by the server's return types, not by a prompt.

Links

• Site: https://flyvolo.ai · How to read it: https://flyvolo.ai/llms.txt
• Method and evidence stages: https://flyvolo.ai/en/method
• Source: https://github.com/weihermans/VOLO (server) · this bridge: https://github.com/weihermans/volo-mcp

MIT.