# Jira's depth, on demand: designing an MCP server around jobs, not endpoints

DevFeed: [Jira's depth, on demand: designing an MCP server around jobs, not endpoints](<https://devfeed.tech/articles/jira-s-depth-on-demand-designing-an-mcp-server-around-jobs-not-endpoints-105639.md>)

Original publisher: [Read original article](<https://www.atlassian.com/blog/how-we-build/jira-depth-on-demand>)

Author: Jovana Dunisijevic

Published: 2026-10-28T21:52:28Z

Content type: article

Language: en

Sources: [How We Build Archives - Inside Atlassian](<https://devfeed.tech/sources/how-we-build-archives-inside-atlassian.md>)

Topics: [jira](<https://devfeed.tech/topics/jira.md>), [MSP MCP](<https://devfeed.tech/topics/msp-mcp.md>), [software-development](<https://devfeed.tech/topics/software-development.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [how-we-build](<https://devfeed.tech/tags/how-we-build.md>), [jira](<https://devfeed.tech/tags/jira.md>), [jobs](<https://devfeed.tech/tags/jobs.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [mcp-server](<https://devfeed.tech/tags/mcp-server.md>), [server](<https://devfeed.tech/tags/server.md>)

## AI overview

Atlassian's Jira MCP v2 groups tools around user jobs so the server coordinates multiple Jira API calls while the agent focuses on outcomes. Its 52 tools cover more than 100 APIs; six are always available and 46 are deferred through discovery and risk-tiered execution. The design uses intent-based tools, action enums, name-based lookups, actionable errors, and JQL and detail-level options. Workflow checks still apply: closing an issue may require first listing valid transitions and required fields. OAuth 2.1 preserves the signed-in user's permissions, while admins can separately scope destructive and administrative tools; Atlassian Guard policies can restrict MCP access to projects or sensitive content. In an internal 84-case October evaluation, v2 passed 57 of 62 tasks both versions could attempt, versus 54 for v1, and reduced failed calls from 0.20 to 0.09 per run. v2 also completed 17 cases unsupported by v1. Internal benchmarks report up to 25% fewer tokens for the same Jira work; a separate comparison found loading the full catalog used roughly 2.2 times the cumulative input tokens of deferred loading. Some evaluation cases still fail, including one where the returned backlog,

## Source excerpt

In this blog we're going to walk you through our Jira MCP tool design journey, a critical part of Atlassian's MCP AI capability, supporting more than 15M calls every weekday. "MCP on Jira is the greatest advance since the airplane." - real quote from a happy Jira MCP customer Ask an agent to close a [...] The post Jira's depth, on demand: designing an MCP server around jobs, not endpoints appeared first on Inside Atlassian.