Based on the README, this is an officially maintained, broad-coverage MCP server for the Harness platform with detailed configuration and security documentation, though its write-capable tools carry real operational risk and should be deployed with careful auth and read-only settings.
Strongest · Documentation 17/20 Weakest · Reliability 8/20
Static review · not runListed 2026-08-26
Read the FMRS scoring method →Runtime：Node.js · Docker
HARNESS_API_KEY
Harness personal access token or service account token in the format pat.<accountId>.<tokenId>.<secret>; create it under My Profile → API Keys in the Harness web UI.
HARNESS_MCP_AUTH_TOKEN
Bearer token required on /mcp HTTP routes when set; mandatory by default for remote HTTP deployments; generate it yourself.
HARNESS_FME_API_KEY
Optional FME/Split Admin credential for fme_ resources; from the Split/FME console or an FME-entitled Harness token.
HARNESS_SEARCH_SERVICE_HEADERS
JSON object of headers sent to the remote search service, supporting any auth scheme; treat as secret when it carries credentials.
HARNESS_AUDIT_WEBHOOK_TOKEN
Optional bearer token sent to the audit webhook.
HARNESS_ACCOUNT_ID
Harness account ID; usually auto-extracted from PAT/SAT tokens that embed it.
HARNESS_BASE_URL
Harness API/UI base URL, defaults to https://app.harness.io; set for private environments such as harness0.
HARNESS_ORG
Default org ID used when org_id is not passed per tool call; optional.
HARNESS_PROJECT
Default project ID used when project_id is not passed per tool call; optional.
HARNESS_MCP_MODE
Deployment mode: single-user (default) or multi-user (per-session credentials over HTTP).
HARNESS_READ_ONLY
Set to true to block all mutating operations; only list and get are allowed, useful for shared/demo environments.
HARNESS_TOOLSETS
Comma-separated toolset filter list controlling which resource categories are exposed; default toolsets load when empty.
HARNESS_FME_BASE_URL
Split/FME Admin API base URL, defaults to https://api.split.io.
HARNESS_SEARCH_PROVIDER
Semantic search backend for harness_search: local (default), remote, or none.
HARNESS_SEARCH_SERVICE_URL
Base URL of the remote search service; required when HARNESS_SEARCH_PROVIDER=remote.
HARNESS_AUTO_APPROVE_RISK
Risk-based auto-approval threshold (none/low_write/medium_write/high_write/all) for autonomous workflows.
HARNESS_ALLOW_HTTP
Allows a non-HTTPS HARNESS_BASE_URL; only for local development.
HARNESS_PIPELINE_VERSION
Pipeline YAML version (0 or 1, Alpha) choosing between pipeline and pipeline_v1 resources.
HARNESS_MCP_ALLOWED_HOSTS
Hostnames allowed by HTTP Host-header validation for DNS-rebinding protection.
HARNESS_MCP_ALLOW_UNAUTHENTICATED_HTTP
Explicitly allows unauthenticated HTTP on non-loopback binds; use with caution.
HARNESS_API_TIMEOUT_MS
HTTP request timeout in milliseconds for Harness APIs, default 30000.
HARNESS_MAX_RETRIES
Retry count for transient failures (429/5xx), default 3.
HARNESS_MAX_BODY_SIZE_MB
Max HTTP request body size in MB for http transport, default 10.
HARNESS_RATE_LIMIT_RPS
Client-side request throttle (requests per second) to Harness APIs, default 10.
LOG_LEVEL
Log verbosity: debug/info/warn/error, default info.
HARNESS_MCP_LOG_FILE
File for stdio disconnect/crash diagnostics, default ~/.claude/harness-mcp.log.
HARNESS_LOG_UNSAFE_BODIES
Include raw request/response bodies in logs; off by default since bodies may contain secrets.
HARNESS_AUDIT_FILE
File for appending newline-delimited JSON audit events for local collection.
HARNESS_AUDIT_WEBHOOK_URL
HTTPS endpoint receiving batched audit events.
HARNESS_AUDIT_WEBHOOK_BATCH_SIZE
Number of audit events batched before webhook flush, default 10.
HARNESS_AUDIT_WEBHOOK_FLUSH_MS
Max time to hold audit events before webhook flush, default 5000.
OTEL_EXPORTER_OTLP_ENDPOINT
When set, enables OpenTelemetry audit spans (optional OTel packages must be installed).
HARNESS_HF_CACHE_DIR
Cache directory for the HuggingFace model used by the local search provider, default /tmp/hf-cache.
HARNESS_DIAGNOSE_LOG_FETCH_CONCURRENCY
Max concurrent log-blob downloads issued by harness_diagnose, default 3.
HARNESS_MCP_TRUST_PROXY
Number of reverse proxy hops to trust for client IP resolution, default 0.
HARNESS_SKIP_ELICITATION
Deprecated — use HARNESS_AUTO_APPROVE_RISK=all instead; kept for backward compatibility.
HARNESS_API_KEY=pat.xxx.xxx.xxx npx harness-mcp-v2@latest; alternatively npm install -g harness-mcp-v2 for a global install, or clone the repo and build from source with pnpm (pnpm install && pnpm build).command pointing to the absolute path of npx or the installed harness-mcp-v2 binary, and set HARNESS_API_KEY in env (GUI clients also need an explicit PATH including the node directory to avoid npx ENOENT errors).npx harness-mcp-v2 http --port 8080 and configure security-related env vars like HARNESS_MCP_AUTH_TOKEN.{"mcpServers":{"harness":{"command":"/absolute/path/to/npx","args":["-y","harness-mcp-v2@latest"],"env":{"HARNESS_API_KEY":"pat.xxx.xxx.xxx","PATH":"/directory/containing/node:/usr/local/bin:/usr/bin:/bin"}}}}
Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.
{
"servers": {
"harness": {
"command": "/absolute/path/to/npx",
"args": [
"-y",
"harness-mcp-v2@latest"
],
"env": {
"HARNESS_API_KEY": "pat.xxx.xxx.xxx",
"PATH": "/directory/containing/node:/usr/local/bin:/usr/bin:/bin"
}
}
}
}
Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).
claude mcp add harness -e HARNESS_API_KEY=pat.xxx.xxx.xxx -e PATH=/directory/containing/node:/usr/local/bin:/usr/bin:/bin -- /absolute/path/to/npx -y harness-mcp-v2@latest
Run it in a terminal; replace any <…> placeholders with your own values first.
After adding HARNESS_API_KEY to the client config and restarting, the client's tool list should show the 11 harness_ tools (e.g. harness_list, harness_execute). Ask "List the organizations in my account" — a returned org list confirms the connection works.
Once connected, you can ask your AI assistant things like:
No matching tools
Listed from the project's documentation, not tested by this site.
Harness MCP Server 2.0 is the official Model Context Protocol server maintained by Harness, giving AI agents full access to the Harness.io platform. Instead of mapping one tool per API endpoint, it uses a registry-based dispatch system with 11 consolidated tools (harness_list, harness_get, harness_create, etc.) that route to 224 resource types across 38 default toolsets spanning CI/CD, GitOps, Feature Flags, Cloud Cost Management, Security Testing, Chaos Engineering, Database DevOps, Internal Developer Portal, Software Supply Chain, Infrastructure as Code, Governance, and more. It supports zero-install startup via npx, global npm install, or building from source, offers both stdio (local clients) and HTTP (remote/shared) transports, and is ready for Docker and Kubernetes deployment. Only a Harness API key (PAT or SAT) is required, with account ID auto-extracted from the token. Mutating operations (create, update, delete, execute) prompt for user confirmation via elicitation by default, with a configurable risk-based auto-approval threshold.
Upstash's official server providing up-to-date third-party library docs for AI coding assistants
AI-powered NPM package analysis MCP server
An independent control layer that gives AI coding agents budgets, brakes, and receipts
GitHub's official MCP server for managing repos, issues, PRs, and workflows via natural language
Source revision b563de3e95ec Data synced 2026-10-12 Read the FMRS scoring method