Taiga MCP Server

Project Overview

Implements a Starlette-based Model Context Protocol (MCP) server that exposes both Server-Sent Events (SSE) and Streamable HTTP transports for ChatGPT and other MCP compliant clients.
Serves as a bridge between ChatGPT and the Taiga project management API, currently including an echo tool for transport validation with ongoing integration of Taiga-specific actions.
Deployed as a container workload to Azure Container Apps and packaged for distribution through GitHub Container Registry (GHCR).

app.py instantiates FastMCP with the Taiga MCP identity, mounts SSE at /sse/ and Streamable HTTP at /mcp/, and disables trailing-slash redirects to preserve MCP session headers.
A custom middleware rewrites bare /mcp requests to /mcp/ and normalizes blank paths inside the sub-application to avoid 307 redirects that previously dropped MCP session headers.
The FastMCP session manager is started via Starlette's lifespan hook to ensure streamable sessions stay active for long-running ChatGPT conversations.
Health checks:
- GET / → plain text "Taiga MCP up" for quick availability probes.
- GET /healthz → minimal health endpoint for container orchestrators.

/ — Root status page.
/healthz — Liveness probe used by Azure Container Apps.
/sse/ — Server-Sent Events transport; requires Accept: text/event-stream and returns the message posting endpoint in the first event payload.
/sse/sse/messages/ — POST target for SSE message submission (returned in SSE endpoint event).
/mcp/ — Streamable HTTP transport; clients must send Accept: application/json, text/event-stream to satisfy protocol negotiation.

echo(message) — diagnostic helper that returns the provided message.
taiga.projects.list() — lists projects visible to the service account.
taiga.epics.list(project_id) — lists epics for a project, including id/ref/subject/status metadata.
taiga.stories.create(project_id, subject, description?, status?, tags?, assigned_to?) — creates a Taiga user story; status accepts either an id or status name/slug.
taiga.epics.add_user_story(epic_id, user_story_id) — links a user story to an epic.

Prerequisites
- Python 3.11 (project uses a .chat-venv virtual environment by default).
- Docker Desktop for container builds.
- Azure CLI for deployment automation.
- GHCR authentication (docker login ghcr.io).
Install dependencies
- python -m venv .chat-venv
- .\.chat-venv\Scripts\python.exe -m pip install -r requirements.txt
Run the server locally
- .\.chat-venv\Scripts\uvicorn.exe app:app --host 127.0.0.1 --port 8010
- Streamable probe: .\.chat-venv\Scripts\python.exe streamable_client.py http://127.0.0.1:8010/mcp --message "hello local"
SSE manual test
- curl.exe -sN -H "Accept: text/event-stream" http://127.0.0.1:8010/sse/

Build tagged images:
- docker build -t ghcr.io/johnwblack/taiga-mcp:v0.0.10 -t ghcr.io/johnwblack/taiga-mcp:latest .
Push to GHCR:
- docker push ghcr.io/johnwblack/taiga-mcp:v0.0.10
- docker push ghcr.io/johnwblack/taiga-mcp:latest

Resource group: rg-offset3
Container app: taiga-mcp
Managed environment: cae-offset3
Deployment command (after successful image push):
- az containerapp update -g rg-offset3 -n taiga-mcp --image ghcr.io/johnwblack/taiga-mcp:v0.0.10
CLI prerequisites on Windows (prevents permission errors):
- $env:AZURE_EXTENSION_DIR = Join-Path $HOME '.az-extensions'
- $env:AZURE_CONFIG_DIR = Join-Path $HOME '.az-cli'

Secrets are stored in Azure Container Apps and surfaced as environment variables for the MCP process.
Commands used:
- az containerapp secret set --resource-group rg-offset3 --name taiga-mcp --secrets taiga-username="info@offset3.com" taiga-password="<PASSWORD>"
- az containerapp update --resource-group rg-offset3 --name taiga-mcp --set-env-vars TAIGA_USERNAME=secretref:taiga-username TAIGA_PASSWORD=secretref:taiga-password
Environment variables available inside the container:
- TAIGA_BASE_URL — base URL for Taiga API (secret reference taiga-base-url).
- TAIGA_USERNAME — service account username.
- TAIGA_PASSWORD — service account password.
- Legacy variables TAIGA_USERNAME_SECRET / TAIGA_PASSWORD_SECRET remain for backward compatibility.
Rotate passwords regularly and repeat the secret update commands to propagate changes.

Streamable HTTP smoke test:
- .\.chat-venv\Scripts\python.exe streamable_client.py https://taiga-mcp.politeground-c43f6662.eastus.azurecontainerapps.io/mcp --message "ping from retest"
SSE availability test:
- curl.exe -sN -H "Accept: text/event-stream" https://taiga-mcp.politeground-c43f6662.eastus.azurecontainerapps.io/sse/ --max-time 5
Azure logs review:
- az containerapp logs show -g rg-offset3 -n taiga-mcp --tail 50
ChatGPT connector validation:
- Configure ChatGPT with the MCP endpoint https://taiga-mcp.politeground-c43f6662.eastus.azurecontainerapps.io/mcp and confirm that the echo tool responds.
- Verify Taiga API routes once exposed by running list and write operations through ChatGPT.

Not Acceptable: Client must accept text/event-stream — Ensure clients send Accept: application/json, text/event-stream when calling /mcp/.
Session terminated errors usually indicate a redirect; verify the request hits /mcp/ (with trailing slash) and that proxies are not rewriting headers.
Azure CLI WinError 5 permission issues are resolved by setting AZURE_EXTENSION_DIR and AZURE_CONFIG_DIR to user-writable locations.
After updating secrets, Azure Container Apps restarts the revision; allow 1–2 minutes before retesting endpoints.

Refactored Starlette routing to eliminate automatic slash redirects that stripped MCP session headers on Azure.
Added middleware to normalize Streamable HTTP paths and ensure compatibility with Azure ingress rewrite behavior.
Built and published container versions v0.0.4 through v0.0.10, with v0.0.10 deployed as revision taiga-mcp--0000016.
Validated SSE and Streamable HTTP transports using local clients, curl, and the ChatGPT MCP connector.
Stored Taiga service account credentials securely in Azure Container Apps secrets to enable future write access to Taiga.