/mcp-server. Every tool is available to any MCP-compatible agent host. Claude Code, Cursor, Codex, OpenCode, or your own client, using a Bearer token for authentication. Tools are grouped into 16 functional categories covering the full research lifecycle, from creating and branching nodes to leasing GPU compute, exporting subgraphs, and auditing activity.
Calling convention
All requests are HTTP POST to/mcp-server with Content-Type: application/json and an Authorization: Bearer <token> header. The body is a standard JSON-RPC 2.0 envelope with method set to tools/call.
Must be
"2.0".Client-generated request identifier. Echoed back in the response.
Always
"tools/call" for tool invocations.The tool name, e.g.
thesis_stage_node_create.Tool-specific arguments. Schema varies per tool; see the individual tool pages.
Example: create a node
result.content array always contains a single text element whose text field is a JSON-encoded object. Parse that string to access the returned data.
Idempotency
Mutation tools in Thesis are automatically idempotent. When you call a mutation tool, the MCP server computes a SHA-256 key from the canonical JSON serialization of your parameters. If you send the exact same parameters again within 7 days, the server replays the cached response without re-executing the operation. This protects against duplicate writes caused by network retries or agent restarts.Idempotency is keyed on the full parameter set. Changing any argument, even whitespace in a string field, produces a new key and triggers a fresh execution.
Tool categories
The 86 tools are grouped into 16 categories. Mutation tools (those that write or modify state) are marked in each category’s reference page.| Category | Count | What it covers |
|---|---|---|
| Nodes | 15 | Create, update, commit, branch, merge, and query research nodes |
| Sharing | 3 | Set per-node or batch sharing modes and collaborators |
| Tags | 4 | Create, rename, and assign tags to nodes |
| Artifacts | 7 | Upload, retrieve, preview, and delete node file attachments |
| Knowledge | 9 | Index papers, repos, docs, and datasets; search and read sources |
| Oracle | 4 | Queue and stream long-running research synthesis jobs |
| Tracer | 2 | GitHub code search across repositories |
| Document Agent | 1 | Q&A against a single indexed source |
| Compute | 11 | Lease GPU instances and manage approval sessions |
| Workflow | 7 | Start, progress, and complete multi-step research workflows |
| Campaigns | 5 | Manage campaign budgets and snapshots |
| Execution | 3 | Launch, list, and terminate code executions |
| Auth & Meta | 4 | Check auth status, credits, and system contract |
| Export/Import | 6 | Export subgraphs as JSON, markdown, or PDF; import subgraphs |
| Updates | 4 | List and manage in-app notifications |
| Audit | 1 | Retrieve audit events for a node |
- Node and graph tools. Create, branch, commit, merge, tag, share, and upload artifacts to nodes.
- Knowledge and search tools. Index sources, run unified search, launch Oracle jobs, and query documents.
- Compute and workflow tools. Acquire GPU leases, run workflows, manage budgets, and launch executions.
- Export, import, and audit tools. Export subgraphs, generate PDF/markdown summaries, and review audit trails.