Pragma × MCP
Your agents deserve better than a TODO.md.
Pragma is a full MCP server behind a real project graph. Agents plan, decompose, link, run impact analysis and move work through actual workflows — scoped by your consent, attributed in your activity history, and never billed by the token.
Agents don't need another task list. They need structure.
Markdown plans go stale the day they're written. Ticket trackers give an agent a flat list with no way to tell what matters, what blocks what, or why any of it exists. So your agent guesses — confidently.
Pragma gives agents a typed graph instead: tasks answer requirements, requirements answer deliverables, dependencies are real edges. An agent can ask "what can start right now?", "what happens if this slips?", or "which work answers nothing?" — and get computed answers, not vibes. Structure an agent can't hallucinate around: cycles, orphans and gaps are detected, not debated.
And because every agent connects through the same server as the app itself, agents and humans work on one plan — not a copy that drifts.
81 tools. Not 81 wrappers.
Most PM tools expose create-and-list. Pragma's MCP server ships the analysis layer too — and teaches every connected agent its planning method: separate the why (which work answers which promise) from the when (what must finish before what), then validate the plan is actually schedulable.
Graph analysis
23 toolsThe tools no other tracker gives an agent.
graph_get_critical_pathgraph_impact_analysisgraph_find_bottlenecksgraph_find_orphansgraph_detect_cyclesgraph_validate_structuregraph_suggest_linksgraph_get_actionable_items
Plan & build
20 toolsProjects, phases, plans, items and typed links.
create_projectbulk_create_manageablescreate_linkcreate_phasepatch_phase_manageablesupdate_manageable
Find & read
13 toolsSemantic search and structured queries, permission-filtered.
searchquery_manageablesget_project_overviewaggregate_manageables_by_tag
Run the work
25 toolsBoards, workflows, tags, assignment.
kanban_move_cardavailable_transitionstransition_manageableassign_activityattach_tag_to_manageable
Autonomy you can audit.
Consent, scoped at connection
When you connect an AI app (OAuth 2.1, no API keys pasted into config files), you choose what it can touch: which projects, which areas, which item types, which actions. The restriction is enforced on every call — not suggested in a prompt.
Every change signed
Agent actions land in your activity history under the agent's own name. "What did Claude change last week?" is a filter, not a forensic investigation.
Humans keep the gates
Workflow conditions apply to agents exactly as to people: a requirement can't be closed while open work answers it, and assignment can be a request the person accepts — even when an agent asks.
Connect in two minutes
-
1
Create your organization — the free plan includes a connected agent.
-
2
Open Settings → Connect with MCP and copy your server address.
-
3
Add it to your client — for Claude Code:
claude mcp add pragma --transport http <your server address>Your browser opens once to sign in and choose what this client may access. That's it — no keys to manage.