Connect an AI client with MCP
Model Context Protocol (MCP) lets an AI application call tools in other software. An OriginChainDB MCP server can act as the connection between that application and database requests.
Availability: MCP is hidden in the new one-token console flow while a compatible published server is being verified. Earlier setup notes describe a separate early-access implementation. Do not use an old package command or credential template as a current installation guide.
For example, a schema-list tool can return the names of available tables and fields. That helps the client prepare a query; it does not prove the generated query is correct.
- The client selects a toolIt chooses a database operation for the question.
- The server applies credentialsIt uses its configured connection and database permissions.
- The database handles the requestThe chosen query model needs its data and indexes prepared.
- The client receives the resultReview returned records before relying on the generated answer.
Before you connect
Contact the team to confirm whether a compatible server is available for your client and deployment. Ask for the package source, supported client versions and the exact credential contract before connecting.
Use a test dataset that you are allowed to share with the AI client and its model provider.
Configure the connection
There is no current public installation command in this guide. Use only the package and configuration supplied for a confirmed compatible release.
The application API now uses the complete address supplied as OC_URL and its supported OC_TOKEN credential. Do not assume that a legacy MCP package accepts that contract or translate older host/tenant settings yourself. Never paste a token into an AI prompt.
Available tools
The earlier implementation described these tools. Their names explain the intended tasks; they are not a claim that a current package is published.
| Tool | Task |
|---|---|
oc_list_schemas | Inspect tables and fields. |
oc_sql | Request supported SQL operations. |
oc_ask | Ask a plain-language database question where enabled. |
oc_vector_topk | Find similar stored vectors. |
oc_fts_search | Find words in indexed text. |
Each search method needs its own prepared data and index. Several tool calls remain separate requests.
Check the setup with a small request
After the team confirms a compatible package:
- Verify that its tools appear in your client.
- Request the available table names and compare them with the intended database.
- Try a small read from known sample data.
- Review the returned data and any tool error before trying another operation.
If setup fails, share the client version, tool name and redacted error with support. Do not include credentials.
Decide what the client can access
Use only the access required for the task. An AI client is not automatically read-only, and a tool call does not bypass database permissions. Review how the client handles results and which model provider receives them.
Read about database permissions or ask about MCP availability.