Off Shoring MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| OFFSHORING_MCP_DB_PATH | No | Where the local story database lives. | var/off-shoring.sqlite3 |
| OFFSHORING_MCP_AUDIENCE | No | Expected audience of the access token. | resource URL |
| OFFSHORING_MCP_JWKS_URI | No | JWKS endpoint for local signature verification. | |
| OFFSHORING_MCP_JWKS_FILE | No | Path to a JWKS file for local signature verification. | |
| OFFSHORING_MCP_JWKS_JSON | No | Inline JWKS document. | |
| OFFSHORING_MCP_LOG_LEVEL | No | debug, info, warning or error. | INFO |
| OFFSHORING_MCP_ISSUER_URL | No | Issuer identifier of the authorization server; must match the token's iss claim. | |
| OFFSHORING_MCP_RESOURCE_URL | No | This resource server's own URL; the default audience. | http://127.0.0.1:8787/mcp |
| OFFSHORING_MCP_REQUIRED_SCOPES | No | Scopes every access token must hold (space- or comma-separated). | |
| OFFSHORING_MCP_ALLOWED_ALGORITHMS | No | Accepted signature algorithms. Never add none or an HS* algorithm. | RS256,RS384,…,EdDSA |
| OFFSHORING_MCP_INTROSPECTION_ENDPOINT | No | RFC 7662 introspection endpoint. | |
| OFFSHORING_MCP_MAX_REQUEST_BODY_BYTES | No | Maximum accepted HTTP request body. | 1000000 |
| OFFSHORING_MCP_INTROSPECTION_CLIENT_ID | No | Client id for introspection. | |
| OFFSHORING_MCP_INTROSPECTION_CLIENT_SECRET | No | Client secret for introspection. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| list_storiesB | Return compact summaries of stored stories (title, ids, revision, latest update, turn and branch counts), newest update first. Read-only. |
| create_storyB | Create a story with a scene setup and editable character assignments. Writes local fiction records: this creates saved state, it does not begin a conversation. If idempotency_key is supplied, a retry with the same key and arguments returns the original result instead of creating a second story. |
| get_story_contextA | Return the selected branch, character assignments, stored profiles, scene setup, summaries, the story revision, and a bounded window of turns. By default the window contains only completed turns; set include_incomplete=true to also see drafts, cancellations and failures. The window reports truncation flags and a cursor for older turns. Read-only: it never changes saved state. |
| read_turnsB | Read one bounded page of turns from a specific branch. Only that branch's own turns and its ancestors' history up to its fork point are visible; sibling branches never leak into the page. Read-only. |
| update_storyA | Change a story's title, scene setup, character assignments, profiles, or summary. This overwrites the fields you pass, so it changes saved fiction; pass expected_revision from your last read to avoid clobbering a concurrent edit. Summaries are editable context and never replace turns. |
| append_turnA | Save one explicit turn: the speaker, the exact text, and a status. This writes saved fiction and is the only way dialogue enters the store. idempotency_key is required, so retrying the same call cannot duplicate the turn (reusing a key with different arguments is refused). expected_revision must match the current story revision. status=completed means settled canon; draft, cancelled and failed are stored but are not canon. The tool confirms the stored turn id and new revision. |
| branch_storyA | Fork the story at a turn, creating a new branch that inherits history only up to that fork point. Nothing is copied or erased: the source branch is left exactly as it was, and later source turns do not appear on the new branch. Use this for a regeneration or a revision instead of overwriting an original turn. |
| select_branchB | Persist which branch is active for a story, so later reads and writes default to it. Requires expected_revision. Nothing about the branches themselves changes. |
| export_storyB | Render a story as JSON or Markdown containing assignments, profiles, summaries, branches, turns and statuses. The document is returned inline in this result; it is not a downloadable file and no filesystem path is accepted or exposed. Exports can contain private dialogue, so treat the returned text as sensitive. |
| delete_storyA | Delete a story's local records: its branches, turns, profiles, summaries and write records. This is destructive and cannot be undone. It requires confirm=true and the current expected_revision. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| offshoring_scene_turn | Guidance for continuing the stored fictional scene: fetch context, respect the stored character assignments and the human's agency, then save settled turns with append_turn. This is guidance only; it does not call any tool by itself. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| off-shoring-guide | How to use the Off Shoring tools in a conversation. |
TDQS
Scored across 10 tools
Most tools have clearly distinct purposes: create/update/delete/list/export story, append turn, read turns, and branch operations are well separated. The only mild overlaps are select_branch vs branch_story (both branch-related) and read_turns vs get_story_context (both read turn data), but their descriptions make the boundaries clear.
All tool names follow a consistent snake_case verb_noun pattern: select_branch, export_story, delete_story, list_stories, create_story, get_story_context, read_turns, update_story, append_turn, branch_story. No mixed conventions or vague standalone verbs.
Ten tools is well-scoped for a story management server with branching and turn history. Each tool covers a distinct operation, and there is no obvious filler or redundant tool.
Core lifecycle is covered: create/list/get/update/delete/export stories, append/read turns, select/branch stories. Minor gaps exist around explicit branch listing or branch deletion and direct turn deletion, but these may be intentional due to the immutable-turn design; agents can work around them via context and export.