goodmem:mcp

GoodMem MCP Tools — Workflow Guide

The GoodMem MCP server exposes tools across multiple API namespaces plus local utilities. Most tools map to a GoodMem REST API endpoint; the utilities (goodmem_configure, goodmem_lookup_model, goodmem_client_info) run locally without contacting the server. Use tools/list to discover exact parameter schemas — this document covers workflow and patterns only.

Setup: The MCP server needs GOODMEM_BASE_URL and GOODMEM_API_KEY. These can be set as environment variables before launch, or configured from chat via goodmem_configure.

TLS: If the server uses self-signed or private CA certificates, set NODE_EXTRA_CA_CERTS=/path/to/rootCA.pem or NODE_TLS_REJECT_UNAUTHORIZED=0 (local dev only) as an environment variable before launch.

---

Typical workflow

1. Configure — call goodmem_configure if credentials weren't set via env vars.
2. Register providers — create embedder, LLM, and optionally reranker. Use goodmem_lookup_model to check the model registry before creating — it auto-infers provider, endpoint, and dimensionality for known models.
3. Create a space — a space binds an embedder and chunking config. Memories ingested into a space are automatically chunked and embedded.
4. Ingest memories — use goodmem_memories_create (single) or goodmem_memories_batch_create (bulk). Supports text, base64-encoded files, or URL references.
5. Wait for processing — after ingestion, memories are processed asynchronously. Poll with goodmem_memories_get until processingStatus is COMPLETED.
6. Retrieve — goodmem_memories_retrieve performs semantic search. Returns NDJSON with ranked results.

---

Namespaces

Embedders (goodmem_embedders_*)

Register and manage embedding models. The four SaaS provider types (OpenAI, Cohere, Voyage, Jina) require credentials. OpenAI-compatible endpoints (Anthropic, Google, Mistral) also require credentials when used under the OPENAI provider type. Omitting credentials for a known SaaS endpoint throws an error before the request is sent.