mcp-agensgraph-common

# mcp-agensgraph-common Shared core for the AgensGraph MCP servers ([cypher](https://github.com/skaiworldwide-oss/agensgraph-ai/tree/main/mcp-agensgraph/servers/mcp-agensgraph-cypher), [memory](https://github.com/skaiworldwide-oss/agensgraph-ai/tree/main/mcp-agensgraph/servers/mcp-agensgraph-memory), [data-modeling](https://github.com/skaiworldwide-oss/agensgraph-ai/tree/main/mcp-agensgraph/servers/mcp-agensgraph-data-modeling)). It exists so the three servers don't each re-implement (and drift on) the same plumbing. ## What's in it | Module | Responsibility | |--------|----------------| | `config` | Merge CLI args + env vars → config dicts (`connection_config`, `transport_config`, `read_controls`, `pool_config`). Standardized env var names. | | `connection` | DSN building, the pool's lifetime, `ensure_graph`, `run_query` and `read_page`. *(needs the `db` extra)* | | `safety` | `quote_identifiers`, which keeps the case of a label a model wrote. | | `results` | A result as JSON a model can read: `rows_of` / `mapping_rows`, `value_sanitize`, `fit_rows`. | | `transport` | One `run_server` for stdio / Streamable HTTP / SSE with CORS + TrustedHost middleware. | Nothing is re-exported from the package root; each server imports the module it needs, which is what keeps the DB-less data-modeling server from importing psycopg to read a namespace out of the environment. ## What the pool is told once The graph to read, how long a statement may take, and how many connections there may be. All three were per call before, and the first two were a round trip each: a read tool call is 5.00 client-to-server flushes now against 6.00. The graph has to exist before the pool opens, since each connection selects it as it is made — `ensure_graph(dsn, name)` is that, on a connection of its own. `AGENSGRAPH_POOL_MIN_SIZE` and `AGENSGRAPH_POOL_MAX_SIZE` are how the size is found. Both ends are given values on purpose: a pool told only the lower one is that wide at the top as well, and…