Procore MCP Server
Procore MCP 서버
Procore REST API 전체를 Claude와 같은 AI 어시스턴트에 노출하는 MCP 서버입니다. TypeScript와 Model Context Protocol SDK로 구축되었습니다.
Claude Desktop, Claude Code, 그리고 모든 MCP 호환 클라이언트에서 작동합니다.
기능
빌드 타임 파서가 Procore의 OpenAPI 스펙을 컴팩트한 카탈로그로 변환한 후, 모든 API 작업에 대해 개별 MCP 도구를 자동 생성합니다. 런타임에서는 7개의 메타 도구를 통해 AI가 모든 Procore 엔드포인트를 발견하고 호출할 수 있습니다:
도구 | 용도 |
| 엔드포인트 수와 함께 API 카테고리 나열 |
| 카테고리 내 엔드포인트 나열 |
| 모든 엔드포인트 전체 텍스트 검색 |
| 엔드포인트의 전체 파라미터 스키마 가져오기 |
| 모든 Procore API 호출 실행 |
| 현재 설정 및 인증 상태 표시 |
| 런타임 설정 지정 (company_id, project_id) |
Related MCP server: mcp-clickup
사전 요구 사항
Node.js 18+
Authorization Code 권한 유형의 OAuth 앱
리다이렉트 URI를
http://localhost로 설정
설정
git clone https://github.com/TylerIlunga/procore-mcp-server.git
cd procore-mcp-server
npm install예제 env 파일을 복사하고 자격 증명을 입력하세요:
cp .env.example .envPROCORE_CLIENT_ID=your_client_id
PROCORE_CLIENT_SECRET=your_client_secret
PROCORE_COMPANY_ID=your_company_id기본적으로 서버는 7개의 발견 도구를 노출하며, 모든 Procore 엔드포인트는 procore_api_call을 통해 접근합니다. 엔드포인트별 전용 도구를 등록하면 약 4.7MB(~1.2M 토큰)의 도구 정의가 생성되는데, 이는 현재 어떤 모델의 컨텍스트 창보다 크므로 해당 표면은 옵트인 방식입니다:
PROCORE_TOOL_MODE=all두 모드 모두 적용 범위는 동일하며, 광고되는 도구 목록의 크기만 다릅니다. all로 전환하고 v2.0.0 이전 버전에서 마이그레이션하는 경우, 이전 -> 새 도구 이름 매핑은 data/tool-renames.json을 참조하세요.
Procore의 OpenAPI 스펙 파일을 specs/combined_OAS.json에 배치해야 합니다. 이 파일은 크기(~54MB) 때문에 저장소에 포함되어 있지 않습니다. Procore의 API 문서에서 얻을 수 있습니다.
카탈로그를 빌드하고 TypeScript를 컴파일하세요:
npm run buildProcore로 인증합니다 (OAuth용 브라우저가 열립니다):
npm run auth서버를 시작합니다:
npm startClaude Desktop 설정
Claude Desktop 설정(~/Library/Application Support/Claude/claude_desktop_config.json)에 추가하세요:
{
"mcpServers": {
"procore": {
"command": "node",
"args": ["/absolute/path/to/procore-mcp-server/dist/src/index.js"],
"env": {
"PROCORE_CLIENT_ID": "your_client_id",
"PROCORE_CLIENT_SECRET": "your_client_secret",
"PROCORE_COMPANY_ID": "your_company_id"
}
}
}
}Claude Code 설정
프로젝트 루트의 .mcp.json에 추가하세요:
{
"mcpServers": {
"procore": {
"command": "node",
"args": ["/absolute/path/to/procore-mcp-server/dist/src/index.js"],
"env": {
"PROCORE_CLIENT_ID": "your_client_id",
"PROCORE_CLIENT_SECRET": "your_client_secret",
"PROCORE_COMPANY_ID": "your_company_id"
}
}
}
}프로젝트 구조
src/
auth/ OAuth token exchange, refresh, storage
api/ HTTP client with auth, rate limits, retries
catalog/ Endpoint catalog loading, search, filtering
tools/ MCP tool handlers and registration
scripts/
generate-catalog.ts Parse OAS into catalog
generate-tools-manifest.ts Generate per-endpoint MCP tools
validate-catalog.ts Validate catalog integrity
data/ Build output (committed): catalog.json, endpoint details, tools manifest
specs/ Source OAS file (gitignored — too large for the repo)작동 방식
빌드 타임:
scripts/generate-catalog.ts가 ~54MB의 Procore OpenAPI 스펙(3,155개 작업)을 파싱하여 컴팩트한data/catalog.json과 개별 엔드포인트 상세 파일을data/endpoint-details/에 생성합니다. 그런 다음scripts/generate-tools-manifest.ts가 각 API 작업당 하나의 명명된 MCP 도구가 포함된 도구 매니페스트를 생성합니다 — 동일한 경로의 이전 버전 중복을 통합한 후 총 2,929개입니다.각 생성된 도구는 대상, 사용 시점, 먼저 해결해야 할 상위 ID, 반환 값, 실패 방식에 대한 구조화된 설명을 포함합니다. Procore가 더 이상 사용하지 않는 엔드포인트는 설명에 지원 종료 날짜와
(Deprecated)제목으로 등록됩니다. 대화형/oauth/*엔드포인트는 도구로 등록되지 않지만 —npm run auth가 해당 흐름을 담당합니다 —procore_api_call을 통해 계속 접근할 수 있습니다.인증:
npm run auth를 한 번 실행하여 브라우저에서 OAuth 흐름을 완료하세요. 토큰은~/.procore-mcp/tokens.json에 저장되며 만료 시 자동으로 갱신됩니다.런타임: MCP 서버가 카탈로그를 로드하고 7개의 발견 도구를 등록합니다 (
PROCORE_TOOL_MODE=all일 때는 전체 엔드포인트별 표면도 포함). AI 어시스턴트가 도구를 호출하면 서버가 이를 올바른 Procore API 엔드포인트에 매핑하고, 인증 헤더를 주입하며, 속도 제한과 페이지네이션을 처리하고, 응답을 반환합니다.
영감
매일 Procore를 사용하는 건설 엔지니어인 여자친구를 위해 만들었습니다.
라이선스
MIT
Available Tools
7 toolsprocore_api_callExecute Any Procore API CallADestructive
Executes any Procore REST API call. This is the only tool here that reaches Procore and the only one that can change data — resolve the exact method, path, and parameters with procore_get_endpoint_details first. WRITES ARE REAL: DELETE permanently removes the record, POST creates one, and PATCH/PUT overwrite fields, so confirm the target id before calling and prefer a GET to verify it exists. Handles OAuth from the saved tokens, substitutes {placeholders} from path_params, and rewrites double underscores in query keys into brackets (filters__status becomes filters[status]). company_id and project_id fall back to whatever procore_set_config holds when the path needs them and you omit them. Returns the parsed JSON response together with pagination and rate-limit metadata. Failures come back as an error payload carrying the HTTP status — commonly 401 when the token has expired, 403 without tool permission, 404 when an id does not resolve, 422 when the body fails validation, and 429 when the rate limit is exhausted. Step 4 of the workflow; this reaches every Procore endpoint, including any not exposed as a dedicated tool.
Inputs: path (required), method (required), body, page, +4 more.
procore_discover_categoriesDiscover Procore API CategoriesARead-onlyIdempotent
Lists Procore's API surface as a Category > Module tree with an endpoint count for each. Start here when you do not yet know which part of Procore holds the data you need; if you already know the resource by name ('RFI', 'budget'), procore_search_endpoints gets you there in one step instead of three. The category and module names returned are the exact values procore_discover_endpoints expects. Takes no arguments and returns a JSON object. Reads the catalog bundled with this server, so it makes no Procore request and needs no authentication — it cannot fail with 401/403 and costs no rate limit. Step 1 of the discover -> detail -> call workflow.
Inputs: no named parameters.
procore_discover_endpointsDiscover Endpoints in a CategoryARead-onlyIdempotent
Lists the Procore endpoints inside one category or module, optionally narrowed by a summary substring or HTTP method. Use it after procore_discover_categories to enumerate a focused area; prefer procore_search_endpoints when you have a keyword but no category. Every argument is optional, but omitting all of them returns the entire ~3,100-endpoint catalog, so pass at least a category or a search term. Returns a JSON array of {operationId, summary, method, path}; feed an operationId to procore_get_endpoint_details. Filters that match nothing return an empty array, not an error. Reads the bundled catalog: no Procore request, no authentication, no rate-limit cost. Step 2 of the workflow.
Inputs: module, search, category, method_filter.
procore_get_configShow Server ConfigurationARead-onlyIdempotent
Reports this server's current state: whether Procore OAuth tokens are present and still valid, the default company_id, and the active project_id that procore_api_call substitutes when you omit those parameters. Check this first when a call fails with 401 or 403, or to confirm which project subsequent calls will target before running a write. Never returns token values or the client secret — only whether credentials are present. Takes no arguments and returns a JSON object. Reads local process state, so it makes no Procore request. Pair with procore_set_config to change any of it.
Inputs: no named parameters.
procore_get_endpoint_detailsGet Full Endpoint DetailsARead-onlyIdempotent
Returns the complete parameter schema for one Procore endpoint: every path, query, and body field with its type and required flag, plus the response shape. Call this after discovery and before procore_api_call — api_call needs exact parameter names, and guessing them is the most common cause of a rejected request. Takes the operationId string that procore_discover_endpoints and procore_search_endpoints return; an unrecognized operation_id comes back as a not-found message rather than an error. Reads the bundled catalog: no Procore request, no authentication, no rate-limit cost. Step 3 of the workflow.
Inputs: operation_id (required).
procore_search_endpointsFull-Text Search Across EndpointsARead-onlyIdempotent
Searches every Procore endpoint's summary, tag, and path for a term and returns the matches ranked by relevance. This is the fastest way in when you already know roughly what you want ('punch list', 'submittal', 'budget line item'); reach for procore_discover_categories instead when you want to browse the API surface rather than search it. Returns a JSON array of {operationId, summary, method, path}; feed an operationId to procore_get_endpoint_details to get its parameters. A term with no matches returns an empty array, so retry with a broader or singular form before concluding the endpoint does not exist. Reads the bundled catalog: no Procore request, no authentication, no rate-limit cost.
Inputs: query (required).
procore_set_configSet Runtime Configuration ValueAIdempotent
Sets the default company_id or project_id that later procore_api_call requests use when the path needs one and you omit it. Use it to switch project context mid-session instead of restarting the server, then call procore_get_config to confirm what took effect. Only 'company_id' and 'project_id' are accepted; both are coerced to integers, and a non-numeric value is reported back as a message rather than stored. The change lives in memory for this server process only — it is never written to disk and is lost on restart. Setting the same value twice is a no-op, and nothing in Procore is modified: this only changes which ids this server fills in for you. Returns a confirmation plus the full updated configuration.
Inputs: key (required), value (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 7 tools
Each tool has a clearly distinct role: browse categories, search endpoints, get endpoint details, execute an API call, and read/write config. Even discover_endpoints and search_endpoints both return endpoints, but their descriptions frame them as browse vs. keyword lookup, so an agent should not misselect.
All tools share a procore_ snake_case prefix and most follow a verb_noun pattern (discover_categories, get_config, search_endpoints). The one outlier is procore_api_call, which reads as noun_noun rather than call_api, but it remains clear and does not create real confusion.
Seven tools is well-scoped for a generic API-access server: discovery, search, schema inspection, execution, and config get/set each earn their place. There is no redundancy and no obvious missing companion tool.
The tool set fully covers its stated discover -> detail -> call workflow, with search added for faster lookup and config tools for context switching. Since procore_api_call can reach every Procore endpoint, no dedicated per-resource tools are needed for the server's purpose.
Maintenance
Related MCP Connectors
The official Planning Center MCP server for interacting with your ministry's data.
An MCP server that let you interact with Cycloid.io Internal Development Portal and Platform
The official MCP Server for the Mux API
MCP server for Product Management
Related MCP Servers
- AlicenseNot gradedqualityNot gradedmaintenanceMCP server for Huntress API integration1MIT
- MIT
- AlicenseAqualityAmaintenanceMCP server for Huly (task tracker) integration61,170 npm64MIT
- AlicenseBqualityDmaintenanceMCP Server for Git operations, agent templates, and project utilities.953 npmMIT