Chemia materials MCP connector

What is the Chemia materials MCP connector?

It is a server that follows the Model Context Protocol (MCP), an open protocol that AI assistants use to call tools. It is built for MCP clients that support remote servers, and the Connection guide lists the clients we have tested. Through it, an assistant can call 28 tools (as of 2026-10-04) that search Chemia's materials corpus, read single properties together with the kind of evidence behind them, page through earlier results, and walk a knowledge graph.

The tools fall into these groups:

• Find materials: Search the corpus with a plain-language query, or count how many materials meet some criteria. (2 tools)
• Look up a material: Resolve a name, list every entry for a formula, read one property or all of them, and classify materials as metals, semiconductors or insulators. (5 tools)
• Property distributions and coverage: See how a property is distributed across a dataset, and find routes to a property we do not hold directly. (3 tools)
• Work with a result set: A search returns its matches as a result set. These tools page through it, show what was eliminated, narrow it, rank it, chart it, compare it with an incumbent and export it. (8 tools)
• Explore the knowledge graph: Walk MatKG, the literature-derived knowledge graph: neighbours, paths, applications and substitutes. (7 tools)
• Orient yourself: List the datasets you can query and the capabilities of the wider platform. (2 tools)
• Computed on demand: Values that are not stored and are calculated from outside data when you ask. (1 tool)

Which tools should I try first?

These 8 tools have a full reference page. Each line is the tool's own one-sentence summary.

• search_materials: Finds materials from a plain-language description of the properties and composition you want. Returns up to 50 matching rows, the constraints we understood, how many materials were eliminated, and a result_set_id that the other tools use to page, refine, rank and export the full match.
• count_materials: Returns one exact number: how many materials meet some criteria, or how many the dataset holds when no criteria are given. Use it instead of a search whenever the question is "how many".
• get_material_property: Reads one stored property of one material, with its unit, the kind of evidence it comes from, and the material's structure.
• resolve_material_name: Resolves a commercial or common name, a formula or a Chemia ID to one canonical material, with its Chemia ID, formula and dataset.
• show_eliminated: Shows the candidates an earlier search eliminated, closest miss first, so you can see how near each one came to passing.
• get_result_set: Pages through the stored matches of an earlier search, up to 50 rows at a time, with the match totals and a next_offset to continue from.
• find_property_proxies: When a property is missing or too thin in the corpus, finds ranked routes that reach it through related properties, with each route's equation, stated error, failure modes and the data we actually hold for its inputs.
• list_material_sources: Lists the datasets this server can query, with their ids, descriptions and whether access to them is restricted.

The Tool reference lists every other tool.

What is the connector not?

• Not anonymous. Every call needs an Elixir account that an administrator has approved. There is no anonymous access.
• Not a write interface. Every tool is annotated read-only: 28 of 28 (as of 2026-10-04).
• Not unlimited. No response carries more than 50 rows, and the charged tools share a daily allowance.
• Not a replacement for measurement. A property value says whether it is experimental, computed (DFT), ML-predicted or from the literature. Property proxies rank and screen; they never qualify.

How do I connect?

Add the server URL https://mcp.chemiadiscovery.com/mcp as a custom connector in a client that supports remote MCP servers, sign in with your Elixir account, and approve read access. The Connection guide lists the clients we have tested, with dates, and how to check that the connection works.

Who can use it, and what are the limits?

Only people with an approved Elixir account. The grants on the account decide which tools it sees. The charged tools (compute_pourbaix_stability, count_materials and search_materials) draw on a daily allowance, which is 200 charged calls per day per caller by default, and no response carries more than 50 rows (as of 2026-10-04). Every limit is listed in Access, grants and limits.

How should I read the results?

Each tool carries short reading notes in three classes: Convention, Trap and Gap. As of 2026-10-04 there are 64 notes: 35 Conventions, 21 Traps and 8 Gaps. Reading results explains the classes and lists every note. Read a tool's Traps before you rely on its output.

Where is the machine-readable reference?

Every page on these docs has a Markdown copy at the same address with index.md added. The tool reference is also published as JSON: the same tools, parameters, grants and reading notes that these pages describe.