ABCM MCP Server
OfficialProvides integration with Obsidian vaults, allowing a managed ABCM workspace directory to be used as a vault and supporting vault-to-workspace/project mapping for scoped sync through the Obsidian plugin.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ABCM MCP Servershow me the files in the current workspace"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ABCM MCP Server
ABCM is an MCP and REST server library for managed agent-context workspaces.
The source checkout contains code only. Canonical project documentation is stored in the ABCM workspace castalia-public, project abcm. Its pinned service revision is recorded in abcm-documentation.lock.json.
To materialize the exact documentation snapshot for development or release validation:
ABCM_API_TOKEN=... bun run documentation:exportThe default local endpoints are http://127.0.0.1:8787 for REST and http://127.0.0.1:8787/mcp for MCP. Override the documentation service URL with ABCM_BASE_URL.
For Obsidian, open the managed project directory as a vault:
C:\Users\egor\Documents\Qualia\castalia\документация\Castalia\public\abcmThis direct-folder mode is different from the bidirectional abcm-obsidian-plugin, which maps one vault folder to one server-assigned workspaceId/projectId through the scoped REST sync API. Obsidian device operations are not MCP tools; MCP remains the agent interface. The canonical integration guide is docs/integrations/obsidian.md in the ABCM workspace.
To import an independently managed Obsidian vault, select a separate source directory and enable the documentation overlay:
ABCM_DOCUMENTATION_SOURCE_PATH=/absolute/path/to/selected-vault \
docker compose \
-f deploy/compose.config.yaml \
-f deploy/compose.service.yaml \
-f deploy/compose.obsidian.yaml \
up -dThe selected directory is mounted read-only. It must not equal, contain, or be contained by the canonical workspace directory, including after symbolic links are resolved. Agents cannot provide or change this absolute path through REST or MCP; they operate only on the configured sourceId.
When the selected directory is a legacy corpus that has already been normalized into ABCM, configure fail-closed reconciliation. A workspace-owned manifest pins every sourcePath to one canonical targetPath plus both checksums. adopt-existing records provenance without overwriting the canonical bytes. Missing entries, changed checksums, duplicate targets, and absent targets are conflicts; the fallback import directory is never populated in this mode.
Centralized outcome, feedback, business-evaluation, and task-success APIs are intentionally absent. ABCM does not collect package ratings or model outputs. Repository-local fixtures and Docker benchmark runners remain available to ABCM developers and do not persist data in the service. Documentation lifecycle tools appear only when operator-selected directories are configured.
Every pull request merged into main starts the repository-owned Feature completion context benchmark workflow. It builds an isolated Docker fixture, compares direct file search with ABCM under the ordered relevance, fallback, determinism, isolation, and context-efficiency gates, and retains a machine-readable receipt for 30 days. A failed gate marks the post-merge workflow as failed; operators can rerun the same workflow manually. The benchmark never reads the live ABCM workspace or stores client feedback in the service.
To publish the complete 41-tool runtime, including versioned context-selection/v4 entrypoints and documentation lifecycle for both managed workspaces, use the full-capabilities overlay instead of the documentation-only overlay:
docker compose \
--env-file .env.local \
--env-file .env.full-capabilities \
-f deploy/compose.config.yaml \
-f deploy/compose.service.yaml \
-f deploy/compose.full-capabilities.yaml \
up -d --buildWhen the local service is exposed through OpenAI Secure MCP Tunnel, create its shared Docker network once and include the tunnel overlay on every service replacement:
docker network create abcm-runtime
docker compose \
--env-file .env.local \
--env-file .env.full-capabilities \
-f deploy/compose.config.yaml \
-f deploy/compose.service.yaml \
-f deploy/compose.full-capabilities.yaml \
-f deploy/compose.secure-mcp-tunnel.yaml \
up -d --buildThe overlay keeps the REST container on the tunnel network under the internal name abcm-local and admits that hostname at the MCP Host boundary. Restart tunnel-client after the REST container becomes healthy. Omitting this overlay while replacing the service leaves ChatGPT discovery with a tunnel-side 502 or Forbidden response.
Set ABCM_DOCUMENTATION_PUBLIC_SOURCE_PATH, ABCM_DOCUMENTATION_PRIVATE_SOURCE_PATH, and the separate ABCM_ARTIFACT_AMENDMENT_OPERATOR_TOKEN in an ignored operator-owned environment file. The operator token must differ from the agent API token. Restart a long-lived tunnel or connector after the runtime version or tool configuration changes so it performs a new MCP initialization and tools/list.
ABCM domain failures are completed typed MCP outcomes: structuredContent.error_code is the stable discriminator and the JSON text code must be equal. isError=true remains reserved for schema/protocol and unexpected server failures, which prevents managed connectors from replacing an ABCM domain code with a generic connector code.
Markdown tags are the source of link packages. Use frontmatter such as tags: [architecture, control-plane] or inline tags such as #architecture. On the next ScopeMap scan, ABCM exposes one virtual LinkPackage per normalized tag. File create, update, move, delete, batch apply, and documentation sync therefore update packages without a separate publish/rebase lifecycle. Package identifiers never grant access; every build is bound to its workspace and reauthorizes every selected document.
This local environment is a single-version development runtime, not production. After the replacement container is healthy and bun run runtime:inspect succeeds, remove stopped containers and versioned images of the previous ABCM server iteration. Preserve the current container/image, abcm-state, the tunnel, and the separate abcm-mcp-server:benchmark image. A retained Docker rollback version requires an explicit future parallel-version or production policy.
This server cannot be deployed
Maintenance
Related MCP Connectors
Securely search and manage workspace context files for AI agents and teams.
Browse and manage files in your Moxt AI workspace from any MCP client.
Persistent file storage for AI agents via MCP and curl. Upload, download, and version files.
Your org's AI agents, tasks, runs, search, and brain files as MCP tools and resources.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables file system operations such as listing, reading, and creating files within a scoped local project directory. It provides a secure way to manage local files through standardized MCP tools built with FastMCP.-
- FlicenseNot gradedqualityCmaintenanceExposes a secure, path-confined bridge to a local workspace and git remotes, enabling MCP clients to search, read, write, reset files, and perform git operations.-

Nolane Habitatofficial
FlicenseNot gradedqualityBmaintenanceProvides coding agents with a durable, revision-aware project workspace for semantic context, governed source changes, verification, task checkpoints, and observability through an MCP interface.1-- AlicenseNot gradedqualityBmaintenanceEnables agent clients to safely connect to tools and execution resources through MCP with authorization, approvals, audit, chat-context isolation, SSH/Docker access, and long-running command session tracking.MIT