Freshdesk-MCP is a self-hosted Model Context Protocol (MCP) server whose purpose is to provide Freshdesk customer support data to MCP-compatible AI clients, such as Claude, Cursor, VS Code Copilot, and others.
Main capabilities It can connect AI assistants directly to Freshdesk, supporting ticket search, reading conversation content, extracting text from attachments (including performing OCR on screenshots), and enriching returned results with information such as status names, group names, and company IDs. Features mentioned in the project description include OCR, attachment extraction, and batch summaries.
Available tools and permission notes The material states that find_agent and list_groups use Freshdesk administrator-only endpoints (/api/v2/agents and /api/v2/groups). If the API key belongs to an administrator account, these tools can work normally; if a regular support agent's API key is used, a 403 error will be received.
Prerequisites Using this project requires: a Freshdesk account with API access; a GCP project with the Vision API enabled (optional, only needed for image OCR); and a container runtime, such as Cloud Run, AWS App Runner, Azure Container Apps, or ordinary Docker.
Deployment methods Under Docker (any platform), the image can be built and pushed to Artifact Registry, then deployed. Secrets can be stored in GCP Secret Manager, and the Cloud Run runtime service account can be granted the roles/secretmanager.secretAccessor permission for each secret. The Docker image can also be deployed to AWS App Runner or Azure Container Apps, and configuration variables can be passed in as environment variables or secrets through the platform's secret storage.
Connecting an MCP client After deployment is complete, the configuration can be added to the MCP client's configuration (the material does not provide the specific configuration example content).
Secret rotation When rotating the Freshdesk API key: obtain the new key from Freshdesk → Profile → Profile Settings → API Key; use the command curl -s -u "<new-key>:X" "https://<domain>.freshdesk.com/api/v2/tickets?per_page=1" to verify whether it is available; update the secret storage (for example, GCP Secret Manager: echo -n "<new-key>" | gcloud secrets versions add FRESHDESK_API_KEY --data-file=- --project <project>); redeploy to load the new version. The material notes: never include :X in the stored secret value; that suffix is only used for curl basic authentication syntax.
Local development The server starts on port 8080, and the MCP endpoint is POST /mcp.
Applicable scenarios Suitable for teams that need to let AI clients access Freshdesk tickets, conversations, and attachment content in a self-hosted environment, and want to use OCR to extract text from screenshot-type attachments.
Limitations and information not yet provided The project clearly states that find_agent and list_groups require an administrator API key, and ordinary keys will return 403. Image OCR depends on a GCP project with the Vision API enabled. Information such as charging, pricing, the complete list of specific tools, and the detailed mechanism of batch summaries is not yet provided in the material.
