Observed Signal · May 3, 2026 · Bug Report · Source: DEV Community · Impact: 2/5 · Sentiment: Neutral
Logic Apps MCP: Agent Loop and OAuth Bugs
A developer integrating the Logic Apps MCP server with a BODMAS Agent encountered two reproducible bugs. Bug 1 is an intermittent 'An item with the same key has already been added' BadRequest during Agent Loop tool registration; it is concurrency-sensitive (a 60-request load test showed ~37% overall failure rate and failure rate falls but persists with lower concurrency) and is not covered by the Agent action's default retry policy because it returns 400. Bug 2 is that the built-in MCP connector does not support OAuth for Logic Apps Standard-to-Standard calls: workflows fail to trigger or return 502 gateway errors when OAuth is configured, while direct calls with a bearer token (e.g., Postman) work. Both issues are filed on the Azure Logic Apps GitHub (issue #1526). Recommended interim workarounds include using anonymous or key-based auth in host.json and manual resubmits for failed runs.
Technical bugs in a cloud integration platform (Azure Logic Apps) affect developers building automation and agent workflows; they cause intermittent failures and block OAuth in certain connector scenarios, but are limited in scope and have filed GitHub issues.
Track GitHub Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Intermittent BadRequest 'An item with the same key has already been added. Key: {tool_name}' occurs at Agent Loop tool registration before any MCP call is made.
- Load test across 60 requests showed 38 successes and 22 failures (~37% failure rate); failure rate reduced with lower concurrency but persisted (≈30% even with sequential runs).
- The BadRequest is treated as a client error (400) so the Agent action's default retry policy (which targets 5xx) does not retry, causing terminal failures.
- MCP connector does not support OAuth for Logic Apps Standard-to-Standard scenarios: workflows fail to trigger or return 502; direct calls with a valid bearer token succeed (Postman).
- Both issues are filed on the Azure/Logic Apps GitHub repo (Agent Loop issue: 'An item with the same key has already been added' — issue #1526) and include reproduction steps and workflow JSON.
Connected Companies & Entities
1 Entity mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Agent Tool Calls Can Become Incorrect Real Requests
A technical post describes a production bug where an agent-to-API bridge (used with Azure AI Foundry and an MCP bridge) translated a model tool call into an HTTP request but failed to reliably parse a model-emitted OData filter. The bridge's regex only handled a specific operator, causing date constraints to be dropped while a raw filter string remained in the request body. The article also documents silent failure modes from using asyncio.gather(return_exceptions=True), insufficient logging and metrics, and recommends rejecting or surfacing almost-correct free-form inputs at the edge and asserting contract shape before forwarding requests.
Agent-verification platform recorded false successes
A developer postmortem describing bugs found while building AiOps Enabler, a platform that verifies AI agents' performance. Key failures included a generated GitHub Actions workflow that always reported success on a cron schedule, an OIDC binding keyed to repo plus workflow filename that broke reporting when workflows were consolidated, a scoring curve that miscommunicates a high-performing agent as low (e.g., 44/100 despite 100% success), and a CI gating bug that prevented a merged feature from deploying to production. The author outlines architecture choices, the current product surface (SDKs, API, directory), and lessons about verification, testing, and distribution. The article was published 2026-08-27.
Activepieces and MCP: Building AI Workflows
A 2026 developer post documents a team’s evaluation of Activepieces (an open-source automation platform) for building AI-agent workflows using MCP (Model Context Protocol) servers. The team implemented pipelines that read CRM records, performed web lookups and LLM reasoning, and wrote results back to HubSpot. They encountered three primary failure modes: ambiguous system prompts suppressed LLM confidence scores, Activepieces does not allow a scheduled trigger and a webhook trigger in the same workflow, and silent data loss from a batch node when static persistence was not enabled. The piece contrasts self-hosting advantages (no per-task pricing, auditability) with operational costs and maturity risks in the MCP ecosystem (400+ servers but some unmaintained). It offers actionable takeaways: make LLM scoring rules explicit, separate scheduled and event triggers, measure token consumption early, verify MCP server health, and use non-blocking external writes.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.