MCP Tools Reference¶

0dai exposes its data and capabilities via a Model Context Protocol server. Start the server with 0dai serve, then configure your AI agent to connect to it.

Tools are grouped by whether they read data, write data, or perform system operations.

Read Tools¶

These tools return data without modifying state. Agents should call them at session start to load context.

Tool	Description
`get_project_manifest`	Returns the full contents of `ai/manifest.json`, including project identity, stack, and team settings.
`get_codebase_map`	Returns a structural map of the codebase: directories, key files, and their relationships.
`get_commands`	Returns all commands defined in the manifest (build, test, lint, deploy, etc.) so agents know how to run the project.
`get_discovery`	Returns the auto-detected stack and tooling summary for the current project.
`get_project_health`	Returns a health score and list of issues for the current project based on manifest completeness and sync state.
`get_ai_version`	Returns the current 0dai CLI version and server protocol version.
`get_personas`	Returns all configured agent personas, including role definitions and behavioral overrides.
`get_bulletins`	Returns all active team bulletins — time-sensitive announcements that agents should surface to users.
`get_custom_stacks`	Returns any custom stack definitions registered in the project or org.
`get_applied_lock`	Returns the current applied lock state, indicating whether the manifest is locked against mutations.
`get_environment`	Returns environment context: detected CI system, OS, shell, and relevant environment variables.
`get_telemetry_summary`	Returns an aggregated summary of agent usage telemetry for the current project.
`get_registry`	Returns the current manifest registry state, including remote manifest versions and sync status.
`get_maturity_score`	Returns the project's AI-readiness maturity score across multiple dimensions (documentation, testing, security, etc.).
`search_experience`	Searches the harvested knowledge base for entries relevant to a given query string.
`get_federation`	Returns the federation configuration, showing how this project's manifest relates to other projects in the org.
`get_audit_log`	Returns recent entries from the immutable audit log for the current project.
`get_agent_teams`	Returns all agent team definitions installed in the project, including member roles and responsibilities.
`get_orchestration`	Returns the current orchestration plan if a multi-agent task is in progress.
`get_observability`	Returns observability data for recent agent sessions: latency, token usage, and error rates.
`get_prompt_history`	Returns recent prompt history for the current project, grouped by agent and session.
`get_wal`	Returns recent entries from the write-ahead log, showing pending and applied manifest mutations.
`get_activity_feed`	Returns the team activity feed: recent syncs, promotions, approvals, and session completions.
`get_role_policy`	Returns the role policy definitions that govern what each agent role is permitted to do.
`check_conflicts`	Checks for manifest conflicts between the local state and the remote registry.
`get_knowledge_base`	Returns the full contents of the team knowledge base, including decisions and shared context.
`get_compliance_report`	Returns the most recent compliance report for the current project.
`get_plugins`	Returns all installed 0dai plugins and their registered capabilities.
`scan_secrets`	Scans tracked files for sensitive data patterns (API keys, tokens, credentials) and returns any findings.
`check_approval`	Checks whether a pending manifest change has been approved by required reviewers.
`list_projects`	Returns a list of all projects registered under the current team or org.
`get_project_health_multi`	Returns health scores for multiple projects in a single call — useful for dashboard views.
`get_session`	Returns the current active session from `ai/sessions/active.json`, including goal, plan, files touched, and handoff notes.

Write Tools¶

These tools modify state. Agents should use them to record progress and decisions during a session.

Tool	Description
`create_spec`	Creates or updates a project spec entry in the manifest based on provided requirements or descriptions.
`record_experience`	Stores a new experience entry (a pattern, solution, or lesson learned) in the project knowledge base.
`update_decision`	Records or updates an architectural decision, including rationale and alternatives considered.
`save_session`	Writes the current session state to `ai/sessions/active.json`, enabling handoff to another agent.

System Tools¶

These tools perform operations that affect agent behavior or candidate evaluation.

Tool	Description
`score_candidates`	Evaluates a set of candidate approaches against the project's manifest constraints and returns a ranked score.
`undo_mutation`	Rolls back the most recent manifest mutation using the write-ahead log, restoring the previous state.

Connecting an Agent to the MCP Server¶

Start the server:

0dai serve

The server listens on http://localhost:3000 by default. Configure your agent to use it:

Claude Code (CLAUDE.md):

MCP server: http://localhost:3000

Codex (AGENTS.md):

tools:
  mcp_server: http://localhost:3000

Once connected, agents can call any tool listed above during a session. The get_project_manifest, get_commands, and get_session tools are typically called automatically at session start via the SessionStart hook.

Error Codes¶

Code	Meaning
`NOT_FOUND`	The requested resource does not exist (e.g., no active session).
`LOCKED`	The manifest is locked and the requested write operation is not permitted.
`AUTH_REQUIRED`	The tool requires an authenticated session. Run `0dai auth login`.
`SCHEMA_ERROR`	The input to a write tool failed schema validation.
`CONFLICT`	The write operation conflicts with a concurrent change. Use `undo_mutation` or `check_conflicts` to resolve.