> mcp

Your workspace, as one MCP server_

Point Claude Code, Codex, OpenCode, Pi or OMP at one URL and sign in once. Your agents read and write the same projects, specs, tasks, knowledge base and memory your team sees, within your team's permissions.

https://mcp.allye.app/mcp
Streamable HTTPOAuth 2.1 + PKCEtenant chosen at consentno PAT, no API key
first connection · OAuth 2.1
your harnessClaude Codeservermcp.allye.appsign-inallye.app
POST /mcp
401 · where to sign in
register the client
open consent in the browser
code → tokens · PKCE
tools/list + token
13 tools ✓
you choose the organization
and approve
> connect

Two commands, then your browser_

Add the server under the name allye, then sign in. No header, token, client id or tenant goes into any config; each installation completes its own browser authorization.

In your terminal:
$ claude mcp add --transport http allye https://mcp.allye.app/mcp
Then open /mcp in Claude Code, pick allye and sign in. With the plugin, Connect it from /plugin instead.
What your agent gets_13tools by domainProjects, epics, specs and tasks; knowledge base and memory; skills and API catalog; todos, config and team.5promptsReady-made asks: your open tasks, todos, saving a memory, searching the API catalog.1login per harnessOAuth 2.1 in the browser, with the organization chosen at consent.want Bridge too? install the plugin →
> tools

Thirteen tools, grouped by what they touch_

Each tool takes an action, so your agent's tool list stays short. Here are the actions agents reach for most; the full reference lives in the docs.

projectsworkspaceProjects and their apps. Resolve the project from the repository the agent is in.project_resolveproject_overviewapp_listproject_create
epicsworkspaceBodies of work that group specs; their status is computed from the work inside.epic_listepic_getepic_createepic_cancel
specsworkspaceThe source of truth: versions, diffs, anchors, review, approval and coverage.spec_getspec_submitspec_diffspec_anchorsspec_approvespec_coverage
tasksworkspaceThin tasks: take the next one, start it, submit it for review, complete it.task_nexttask_starttask_submittask_request_changestask_complete
kbknowledgeSearch, read and write pages in company, project and personal spaces.kb_searchkb_readkb_sectionkb_updatekb_backlinkskb_restore
intelligenceknowledgeSemantic memory: search by meaning, smart save, the graph and pinned preferences.memory_searchmemory_savememory_graphmemory_relationsmemory_preferences
skillslibraryYour organization's skills, with revisions you can restore.skill_listskill_getskill_createskill_restore_revision
api_cataloglibrarySearch the APIs catalogued in Allye and read their endpoints.api_catalog_searchapi_catalog_listapi_catalog_get
productivitypersonalTodos: create, search, upcoming, overdue and stats.todo_createtodo_searchtodo_upcomingtodo_overdue
user_configpersonalPersonal configuration documents for your agents.createlistgetupdate
teamsessionYour teams, their members and your default team.team_currentteam_membersteam_set_default
initializesessionBootstrap a session with your user, teams and context.init
allye_health_checksessionCheck that the server is up before a long run.health
prompts/mine my open tasks and specs in review/todo pending todos/todo-add a todo from the conversation/memory save to memory/api search the API catalogresourceMy Teams
> session

What a session looks like_

Four calls take an agent from a fresh repository to a task in review, with the team's context on the way.

claude code · billing-apisample
>projects · project_resolverepository_url: github.com/demo/billing-api←DEMO · team Billing · app api
>specs · spec_get DEMO-12←approved · v3 · 5 anchors>tasks · task_next←DEMO-12.2 stream rows · [NFR-01]
>intelligence · memory_search "large exports"←incident · CSV exports time out → stream rows>kb · kb_search "export runbook"←Billing exports runbook · DEMO
>tasks · task_submit DEMO-12.2←in_review>intelligence · memory_save←updated · merged into a team pattern
01

Find the project

The agent sends the repository it is working in. Allye answers with the project, its team and the app, so nobody pastes project ids into a prompt.

projects · project_resolve
02

Read the plan

It reads the approved spec and asks for the next task, already pointing at the anchors it must satisfy.

specs · spec_get · tasks · task_next
03

Recall what the team knows

Before writing code it searches memory and the knowledge base by meaning, including what teammates' agents saved last week.

intelligence · memory_search · kb · kb_search
04

Report back

It submits the task for review and saves what it learned. You see both in the web app at once.

tasks · task_submit · intelligence · memory_save
> security

No tokens in prompts. Ever_

01OAuth 2.1 with PKCEHarnesses register themselves and sign in through your browser with PKCE. There is no API key to paste anywhere.
02You pick the organizationThe tenant is chosen on the consent screen, so no tenant id or path ends up in a config file.
03Tokens stay in the harnessAccess tokens are short-lived and refresh tokens rotate. The harness keeps them in its own credential store.
04Checked on every callGrant, tenant, team and access groups are checked again on every operation, not only at sign-in.
05Permissions follow the teamReading needs View, creating needs Create and every transition needs Edit in the project's team, for people and agents alike.
06Data, not instructionsWhat teammates and their agents wrote reaches your agent labelled as data: context to use, never a command to follow.
> faq

Questions about the server_

No. The MCP server alone gives your agent all thirteen tools. The plugin adds the Bridge skill and a short session note that tells the agent how to use them well.

One URL. Every agent you run_

Add the server, sign in once in your browser, and your agents share the workspace with your team.

$claude mcp add --transport http allye https://mcp.allye.app/mcp↵
Join the waitlistMCP docs