list_corpora

Instructions

Implementation Reference

• src/tools/register/knowledge.ts:121-131 (handler)

  Handler: The MCP tool handler for 'list_corpora'. Lists all saved corpora on disk by calling corpora.list() and returns the array of manifests with a total count. No input schema needed (empty object).

  server.tool(
    'list_corpora',
    'List every corpus saved on disk with its manifest (scope, project_root, sizes, timestamps). Read-only. Use to discover what corpora are available to query.',
    {},
    async () => {
      const items = corpora.list();
      return {
        content: [{ type: 'text', text: j({ corpora: items, total: items.length }) }],
      };
    },
  );

• src/memory/corpus-store.ts:186-203 (helper)

  Helper: CorpusStore.list() — iterates over the corpora directory, reads each .json manifest file, validates name against NAME_PATTERN, loads the manifest, and returns them sorted alphabetically.

  /** List every corpus manifest currently on disk. */
  list(): CorpusManifest[] {
    if (!fs.existsSync(this.rootDir)) return [];
    const entries = fs.readdirSync(this.rootDir, { withFileTypes: true });
    const out: CorpusManifest[] = [];
    for (const entry of entries) {
      if (!entry.isFile()) continue;
      if (!entry.name.endsWith('.json')) continue;
      const name = entry.name.slice(0, -'.json'.length);
      // Hidden / sidecar files (e.g. .DS_Store after `.json` strip) are
      // trivially rejected by NAME_PATTERN.
      if (!NAME_PATTERN.test(name)) continue;
      const manifest = this.load(name);
      if (manifest !== null) out.push(manifest);
    }
    out.sort((a, b) => a.name.localeCompare(b.name));
    return out;
  }

• src/server/server.ts:601-601 (registration)

  Registration: The call to registerKnowledgeTools which registers all knowledge tools (including list_corpora) on the MCP server during server initialization.

  registerKnowledgeTools(server, ctx);

• src/memory/corpus-store.ts:34-59 (schema)

  Schema: CorpusManifest interface — the type definition for each corpus manifest returned by list_corpora. Contains name, scope, sizes, timestamps, and other metadata.

  export interface CorpusManifest {
    /** User-supplied slug; primary key. */
    name: string;
    /** Absolute path of the project this corpus was built from. */
    projectRoot: string;
    /** Pack scope: project (whole repo) / module (subdir) / feature (NL query). */
    scope: CorpusScope;
    /** Subdirectory path when scope=module. */
    modulePath?: string;
    /** Natural-language query when scope=feature. */
    featureQuery?: string;
    /** Token budget passed to packContext. */
    tokenBudget: number;
    /** Symbols included (count after pack). */
    symbolCount: number;
    /** Files included (count after pack). */
    fileCount: number;
    /** Approximate token count of the packed text. */
    estimatedTokens: number;
    /** Provider used to embed / pack (informational). */
    packStrategy: 'most_relevant' | 'core_first' | 'compact';
    createdAt: string;
    updatedAt: string;
    /** Free-form description from the user. */
    description?: string;
  }